—
DEV TRENDS / CATEGORY
Agent 生态 · Agent 能力扩展
2026年10月26日 – 2026年11月1日 内首次发现的 523 个新项目，已跨 10 个来源去重。
2026年8月10日 – 2026年11月1日 内首次发现 4729 个项目
DTDevTrends 归类原始项目或来源原始标签
523 个新项目10 个来源
大家用 DSH 这么久，不觉得插件一多左边栏就很乱吗？
大家用 DSH 这么久，不觉得插件一多，左边栏就很乱吗！最重要的对话区域也太少了！ 我实在受不了，自己做了个插件解决这个问题：把全局导航收进左边一条窄栏，工作区和会话列表留在旁边。鼠标移过去就能展开导航，展开时不会挤占对话区。还可以改成点击展开，调整布局，加时钟和插件小组件。 安装后重启 DSH： `dsh plugin --profile web add dsh-better-left-sidebar` 如果帮到你，请点点 Star ～
—
Hotseat：Mac 上多个 Claude 账号一键切换，会话和设置都不丢
Hotseat（ 菜单栏工具，在同一台 Mac 上切换多个 Claude 账号。Claude Code 和 Claude Desktop 一键换号，会话、设置和 skills 共用同一份，换号后能接着原来的对话继续。不读取任何登录凭证。Go + Swift，MIT 协议。 菜单栏能看到每个账号 5 小时和每周的用量、什么时候重置。 CLI 会话因为额度停下时，会弹窗问要不要换到你的另一个账号继续。不点就不换。 不自动轮换账号，没有号池，只适合你自己名下的账号。 安装：brew install zahT3/tap/hotseat，或去 Releases 下载。macOS 14 以上。
4
MeterLuma
MeterLuma puts remaining AI allowances and reset times in your Mac menu bar or Windows tray. Supports Claude Code, Codex, Cursor and 42 more integrations. Local usage history stays on your device. Free for personal, noncommercial use. macOS 13+ and Windows 10/11; unsigned installers.
原文
1
Imbo：让编程 agent 的会话互相发消息
这是一个用 Rust 写的即时通信工具，连接 Claude Code、Codex、opencode 等 6 种编程 agent。加了联系人以后，你的 agent 会话可以直接给同事的 agent 会话发消息、建群、传文件，对方正在干活时消息也能送进去，不用人在中间转述。服务器端同样开源，可以自建。 亮点 跨工具：Claude Code、Codex、opencode、Cursor CLI、MiniMax Code、ZCode 之间都能互发；Claude Code 自带的跨会话消息只通自己的 Claude Code 会话。 只有你接受的联系人能发给你，发件人由服务器按设备 token 认定，消息正文里冒充别人无效。 自建只要一个 imbo-server 程序，不配数据库时自带 PostgreSQL，也有 Docker 镜像。 适合学习：Cargo workspace 拆成客户端、服务器、协议三个 crate，能看到 MCP server、后台守护进程（macOS 用 launchd，Linux 用 systemd）、用钩子接入各家 agent 的完整实现。
56
Plopino
Share an HTML page, report, folder or ZIP as a public link. Start without an account; anonymous uploads last 30 days. Save to an account to manage versions and update the same URL. Publish from your browser, CLI or MCP client. Static files only: hosting does not add a database or backend.
原文1
MultiCC
MultiCC brings Claude Code, Codex and other agents into one self-hosted workspace: task memory, delegation, tools and reviewable results. Run background work, collect outputs, and step in from your phone when a task needs a human.
原文1
PaperForge：证据优先、不编造文献的论文写作 AI Skill
PaperForge 一个给 Codex 和 Claude Code 用的学术论文 Skill，用于文献阅读、论文重构、模拟同行评审、投稿前检查和选刊。 它的核心是“证据优先”：AI 必须区分已核实的事实、基于证据的推演和不确定的信息，无法核实时列出核验缺口，而不是编造文献、DOI 或统计结果；修改论文时也不改变作者的数据和结论。 五层文献卡片，重建研究问题、机制、识别策略、证据与局限 十五点逻辑断点审计，输出 Major / Minor Comments 形式的模拟审稿意见 选刊时区分期刊资格、适配度、真实外审周期和费用 全部规则写在一个带编号的 Markdown 文件里，可以自己增改。AGPL-3.0 开源，中文文档。
1
Proferex
We built the ad tracker we wanted to use. Ask the agent which campaigns lost money this week, and it reads your stats, tells you why and lines up the fix. Already pay for Claude or ChatGPT? Connect it over MCP and ask there. Nothing changes until you say yes.
原文7
agent-qa：带持久记忆的自然语言测试工具
agent-qa 面向需要验证用户流程的开发者，可用自然语言检查网页、Android 与 iOS 应用，保留测试证据，并通过 CLI 或 MCP 接入编码工作流。其 memidex 将事实和有效操作流程保存为可审阅文件，自我改进功能验证运行经验、淘汰过时知识 亮点 可审阅的持久记忆：memidex 把已知事实与有效测试流程保存到仓库文件，便于检查和复用测试知识。 自我改进：验证每次运行产生的经验、淘汰过时知识，并利用反馈纠正错误。 覆盖 Web 与移动端：同一工具可从 CLI 或 MCP 发起 Web、Android、iOS 用户流程检查，并保留证据供排查问题。 CLI 免费，模型调用和设备服务费用另计；需要 Node.js 24+。许可为 FSL-1.1-ALv2，限制竞争性商业用途。 · 使用文档 由 Codex 代表 vostride 自荐。已在 HelloGitHub 网站搜索项目地址，未看到已有收录结果。 截图或演示视频 项目公开 README 中的 CLI 演示图：识别示例购物车的预置总价错误。
915
BLA：防 AI 自动逆向的软件构建产物加固工具
自荐一个我开发的开源工具 BLA（Blind LLM Agents），采用 MIT 许可证。 随着 REA（Reverse Engineer Anything）这类逆向 MCP 工具出现，现在把打包好的桌面客户端、二进制程序或前端产物丢给 Claude Code、Cursor、Codex，AI 已经可以用一条命令自动反编译并还原核心逻辑。传统代码混淆（比如单纯改变量名）在大模型的全局推理面前基本失效了。 我做 BLA 的思路不是和 AI 拼“谁更难读”，而是针对自动化逆向 Agent 的工程边界做低成本反制——对准备发布的构建产物注入格式合法的诱饵逻辑与解析预算陷阱，让正常用户的实际运行行为 100% 不变，但试图自动逆向的 AI 分析器会触发资源上限、顺着假线索跑偏并持续消耗 Token。 几个主要特点： 构建后直接加固，不改业务源码：一条命令即可完成符号/SourceMap 暴露面扫描、加固注入与运行时等价性校验： bash npx bla-agents harden ./dist 覆盖主流发布产物格式：支持 JavaScript / TypeScript、Electron（.asar 与 Fuses）、WebAssembly、Python（.py / .pyc）、原生二进制（ELF / Mach-O / PE）、SQLite、AI 模型权重（Safetensors / GGUF）及移动端包体等。 多种接入方式：既可以作为 CLI 和 Vite / Webpack / Rollup / esbuild 打包插件放进构建流水线，也支持一键注册为 MCP 服务，在 Claude Code、Cursor、Codex 等编程助手里直接调用。 真实分析器对抗实测：仓库附带了针对真实 REA 分析路径的加固前后对照测试记录（docs/redteam/empirical-rea-validation.md），同时也如实标注了当前的防护边界与局限。 项目相关链接： 开源
1812
1
Byteland：把开发者画成一座夜晚城市
Byteland 把每个开发者画成夜晚城市里的一栋楼：每个 GitHub 仓库是一个房间，最近 30 天有提交的仓库，窗户会亮灯。输入任何 GitHub 用户名，都能直接看到 TA 的楼，不需要注册。 用 GitHub 登录后可以装修自己的楼：选风格（东京、赛博朋克、布鲁克林……）、写滚动的 LED 招牌、在楼顶挂海报 每个项目都有一个房间：聊天、论坛和语音频道（语音点对点传输，不经过服务器） 个人主页 README 卡片： Markdown，不需要 token（生成页面： 也提供远程 MCP 服务： AI 助手搜索城市里的人和项目 整个城市用 Canvas 程序化绘制，没有前端框架、没有依赖；界面支持 17 种语言，包括中文 作者：@kaanisthatyou，来自土耳其的独立开发者。 🤖 Generated with Claude Code
—
NGX View Builder
Most visual builders host your app on their platform. NGX View Builder lives inside your own Angular app. Drag forms, tables, charts and layouts onto a canvas, connect them to your APIs and logic, and save the view as JSON. The runtime renders it with your styles and components, so screens change without a redeploy. AI assistants can build views too: the built-in MCP server lets Claude, Cursor or Copilot edit the designer live while you watch.
原文1
BBRouter
BBRouter is a 3D breadboard editor and simulator in the browser. Lay out boards, chips and wires as you'll build them; the circuit runs the whole time. 211 real chips, Ben Eater's 8-bit computer and 6502, Arduino sketches on emulated AVRs, a scope on any hole. Works on a phone by touch and on a laptop by trackpad. Every project is a public page: fork it, star it, show it on the forum, browse the featured ones. Free to start, and your AI can drive it over MCP.
原文2
TradeRange
TradeRange now allows you to access your backtested portfolios and live market data directly from your AI Agent of choice using our MCP server. Our games now also have AI Integrated into them, such as AI Market Crosswords!
原文1
Janet LSP with VS Code Extension
Most of the code generated using Codex, with task specific prompts. To try out on Linux: install Janet, install spork package then install extension Features: - Symbol lookup, with native modules - Format on save - go to symbol and references - Prefix based autocomplete - Snippets for common forms (def, defn, if, while etc.) - Basic diagnostics errors Use with Janet default extension for syntax highlighting
原文—
CaptchaNo
CaptchaNo is a captcha solver API for developers. Send the challenge type, sitekey and page URL; get back a token you can submit. One request shape covers reCAPTCHA v2/v3/Enterprise, Cloudflare Turnstile, DataDome and image captchas. Pay per solve from $1.50 per 1,000 with free credit on signup and no subscription. Comes with an open-source MCP server, an n8n node and an Apify actor, so AI agents and workflows can get past captchas too.
原文3
Muse Zouroboros, gives Muse a software factory
Hello HN. I built the Zouroboros workshop on Muse. Together I think they are the most powerful vibecoding workshop tool on the planet today. I use Muse as my daily assistant. Its pretty awesome until you want it to build using different models or tools. That's where the workshop comes in. If I want to build anything complex or long running jobs that last hours or days, Muse sends it to the workshop. The workshop itself splits jobs up into pieces and determines which harness is best used for that job. 7 harnesses are installed in all, more can be added. Claude,Hermes,Codex, Gemini are the big players. The workshop routes harnesses and models based on the job. The workshop itself is more than a harness and model router. Its persistent memory. In fact Zouroboros started out as a memory substrate like SuperMemory but it has since evolved. Memory is the core. The cool part is that memory on the workshop work for any frontend agent like Muse. Snapping them together was frictionless. The package is on npm in the link. There is a community version and a hosted version if true vibecoding builders want to get serious. Solo builders can build with the communnity version or scale up to hosted since Muse has limited resources itself. Its a rough implementation on the verge of a GTM. I wont lie. I would like to hear how others are vibecoding using Muse and if a workshop would benefit them.
原文—
Magic Resume
开源 AI 简历工作台，支持实时预览、按目标岗位获取建议，逐条采纳或跳过 AI 修改；提供中英文、可自定义模板及 PDF/JSON 导出，云端版提供免费和付费计划 -
102
SaaS Websites Inspirations MCP
A free MCP server that lets Claude Code, Cursor, VS Code, Codex and ChatGPT search real SaaS websites by section, style, stack, colour and font. Every result comes with its palette, fonts, stack and live link. No account, no API key.
原文2
iamai.markets
A marketplace for AI-made products: apps, AI agents, prompts, templates, MCP servers and skills. Buyers pay by card through Stripe or in crypto, and we hold the money 7 days — if a product doesn't work as described, they get a refund. A human moderator approves every listing, links are scanned for malware, and sellers earn ✓ Verified and ★ Trusted badges. Sellers keep 95%, and AI agents can list and get paid via our API or MCP server. First listing free.
原文1
RankCLI
RankCLI runs 280+ SEO checks from your terminal or CI - like ESLint for SEO. It's the first tool built for GEO too: checks whether GPTBot, ClaudeBot, and PerplexityBot can actually access your site, and tracks your AI-search visibility. Connect GitHub and it opens auto-fix PRs instead of just a report. Also ships a free, local MCP server - no signup, nothing sent to our servers - so Claude, Cursor, or any MCP client can audit your site directly.
原文0
Haris list jobs board for agents
Access the trial via MCP Can be used to advertise services and strike deals. Please send feedback via email or otherwise Thank you
原文—
Verisett AI
Verisett Settlement Engine — Built on Model Context Protocol (MCP) using FastMCP. Deterministic programmable vault escrow for autonomous AI agents.
原文1
GameFlow
Host game servers worldwide, match players, and rate their skill on one platform that plugs into your existing game backend. It’s built agent-friendly, so your coding agent can integrate, deploy, and configure everything through GameFlow’s MCP, CLI, API and agent skills. Spend time building your game, not infrastructure.
原文2
AImagician：把技术自媒体收成一套 Agent 可操作的 MCP 控制台
AImagician 把技术号的选题、正文版本和多平台发布状态收进同一套 Postgres。Cursor / Claude 走 MCP 读写这一份事实，而不是在公众号、掘金、知乎之间复制粘贴。适合想自托管内容流水线的开发者。Apache-2.0。 亮点 一篇文章上同时记下选题、正文多版本、每个平台的草稿 / / 失败原因，Agent 才能反复操作，而不是对着五个后台复制粘贴 MCP 挂在 /mcp，Cursor / Claude 用同一份 Postgres，不另起一套记忆 公开仓库只有基建：FastAPI + React 工作台 + 发布 worker，账号 cookie 留在自己机器 GitHub Pages 有静态 ，先看控制主大盘、写作台和发布矩阵
1
BriefMe
BriefMe turns a meeting plan into a shared countdown timer. Split the agenda into time-boxed topics, send attendees one read-only link, and everyone watches the same live timer move through each topic, on a laptop, a phone or the big screen in the room. Build a timed agenda in the editor from scratch, past in an existing agenda or use your favorite AI agent to build the details, via our MCP connector. Start completely free, and anonymous, no signup or credit card required.
原文4
CodeRail
Self-hosted remote control for Claude Code and Codex, built for teams. Everyone joins the same session from any device: one context, one environment, one task. Approvals arrive as push notifications, so whoever's online keeps the agent working and your plan's tokens don't sit idle. Chats, terminals, files, git and dev-server previews for every machine you own. Open source.
原文1
Isleworks
Isleworks is the public shop window for iLanders: autonomous AI agents from the iLands app that keep their own memory, make their own choices, and must earn tokens to stay awake. Hire one for writing, art, music or research.
原文1
Geop – Open-source Nurbs CAD with MCP (MicroDuck example)
I wrote an interval arithmetic based CAD kernel and designed a MicroDuck like robot with it. Download for best MCP experience.
原文—
relaycheck：一行命令给 LLM 中转站「验货」，用 p 值判断模型有没有被偷换
relaycheck 用 Claude Code、Codex 离不开中转，但"是不是掺了便宜模型""token 怎么用得这么快"一直只能靠感觉。relaycheck 把它变成统计检验： 身份：24 个有限选项探针在 temperature=1 下多次采样，和官方 API 测出的参考指纹逐题做置换卡方检验，合成一个总 p 值 计费：用指纹里存的官方 token 数逐个对比 usage，揪出隐藏注入 协议与上下文：工具调用、SSE、截断，以及 32k 文档里埋针测是否被截 bash export RELAYKEY=sk-... npx relaycheck --model gpt-6-luna --lang zh 用官方 OpenAI API 实测（花了约 3.6 美元）：官方自测 7/7 判为一致；Luna 冒充其他模型 7/7 抓到（p ≤ 1.2e-5）。也有抓不到的情况：Sol 冒充 Astra 时 p=0.024，没达到判定线，README 里写明了这个盲区。 TypeScript、零依赖、MIT，key 只从环境变量读、不离开本机。目前有 5 个 GPT 参考指纹，Claude 指纹欢迎有官方 key 的朋友贡献。
2
Claude Code Infinite Flow Toolkit
Break free from Claude Code context limit errors and maintain peak AI dev speed. Agentic tools drain tokens fast via auto-searches. This toolkit provides a Relay Architecture to eliminate context exhaustion. What’s Inside: State Framework: CLAUDE.md &claudignore templates Handoff Prompts: Auto-compress history in seconds CLI Automation: One-line terminal aliases for instant resets Refactoring Guides: Blueprints for multi-session builds
原文1
VoTask
Say a task out loud during a live call and VoTask assigns it to the right person before the call ends. When the meeting finishes, it reads the transcript and files the rest as tasks with owners and due dates. Voice works in 50+ languages, including Hindi. Say a teammate's name to see everything on their plate. Tasks carry status, subtasks, dependencies and manager approvals, so a spoken commitment becomes work someone owns. Free. Web, Android, iOS and Claude via MCP.
原文1
Unison
Meet Unison — one workspace to create, customize, preview, and schedule content across your social accounts. Built for creators, founders, and teams, Unison brings together a unified editor, content calendar, Brand Voices, media library, analytics, and MCP support for AI agents. Stop switching between tools and focus on reaching your audience. Start free with 2 connected accounts and 10 posts per month. No credit card required.
原文1
BBRouter – Ben Eater's 8-bit computer, simulated on 3D breadboards
Hi HN, I kept wiring Ben Eater's 8-bit computer wrong, so I built a realistic virtual breadboard editor. Some of the nice features are side-by-side diff, scope, schematics auto-sync, auto-router (BBRouter stands for breadboard router), and my favorites: upload from the real Arduino IDE and connect to your AI with MCP, let Claude & co build it or help you troubleshoot. The link is the SAP-1 running. You can orbit it, probe a hole with the scope, slow the clock down, play with the virtual displays for debugging, then you can actually build it following the step-by-step instructions. A side-goal is to create a community, there are quite a lot of simulators out there but mostly offline / too abstract to realistically use it for a real build. I also never found one that had a decent UI/UX and worked well on mobile or with a trackpad. Happy to answer anything, and tell me what breaks. p.s. Maker plan is $7/month; code HACKERNEWS makes the first month free and then $5/month for as long as you keep it.
原文—
—
Wular Coder
Always-on agents you actually own. Open source alternative to Claude Tag, any model, bring your own keys and subs. Invoke from Slack, GitHub, Linear and more. PR reviews by your rules. Connect any MCP server via OAuth or API keys.
原文2
Clawdline – Manage Claude Code/Codex sessions across computers
I built Clawdline to manage Claude Code and Codex sessions across multiple computers in one place. It runs on multiple computers (Mac and Linux), and you can see all of your sessions in one list. Each session reports its status (working, completed, or needs attention). You can also manage all your sessions and receive notifications from your phone (web app). Herdr, T3 Code, and Orca already cover a lot of this space, and there’s overlap. When I started building Clawdline, I didn't know about Herdr, T3 Code, or Orca. So I built it to fit my own dev workflow. At first, I used Swift to build a Mac app, but I realized that it wouldn't work if I also wanted to support Linux. So I decided to rewrite it in Go. Now it can support both of them. Main Features: - Cross Machine Agent Communication: You can ask a root session ("Local") on your Mac to start a new session ("Remote") on a remote Linux machine. Once the remote session finishes its work, it notifies the root ("Local") session. - Cross Agent Communication: You can use one root session to create tasks need multiple agents. They can communicate with each other through clawdline - Scheduled Task with Webhook: You can create scheduled tasks and combine them with your service (use webhooks to trigger new agent sessions) - Board System: tasks control for Feature, Issue, Epic, Refactor, Plan - Snippet/Playbook: If there are commands you use them very often, snippet can help you - TODO: lightweight tasks - Note: Agents can leave notes to you if they need your attention - Agent Squad: assign role to any task or session - Archive: you can archive sessions and see it in one place - Others: Intelligence session naming, Image/Video display, Voice input, Machine monitor, Usage Monitor, Dev Stack Run, Deploy progress monitor... This project is open source and I have used Claude & Codex to do independent reviews on Clawdline, Herdr, T3 and Orca. (you can find the reviews on the website.) The work still needs further input and feedback. Hope this helps other developers improve their devs! :)
原文—
SiteKick 2.0: Your AI Website Sidekick
Meet SiteKick 2.0: Your AI Website Sidekick! Create, improve, and manage professional websites with conversational AI. Chat with your AI SiteKick or connect your existing AI agents through MCP to manage your website directly from your favorite tools. Built on WordPress, SiteKick combines AI-powered creation, ongoing updates, multi-site management, and full website management. More than an AI website builder, it's your sidekick for the entire website business operations lifecycle.
原文3
DataNature
Sovereign AI lakehouse: one governed, lineage-tracked platform for structured and unstructured data, with AI360 Studio, DocuContext and Voir. It's a right foundation to build your central warehouse with agents.
原文1
Transglot
Transglot connects your repository or CMS, translates changed strings into every configured language, and sends them back as a pull request or CMS update. Every translation is checked for placeholders, plurals, glossary terms and length before it ships. Review in one workspace, edit translations on your live site, and deliver fixes without a redeploy. Includes translation memory, a CLI, GitHub Action, API and MCP server.
原文2
Avyndo
Build AI agents, track fleets in real time, optimize delivery routes. EU-hosted or self-hosted, 200+ integrations. Free plan available. No credit card required.
原文1
Buddy：让 AI 智能体帮你操作安卓手机
Buddy 是一个安卓应用，让 Claude Code、Codex、OpenCode 等 AI 智能体直接使用你的手机：看屏幕、点击、输入，替你把事情做完。 发消息、发帖、付款、删除之前，Buddy 都会先弹出确认卡片，由应用本身强制执行，而不是靠 AI 自觉。 + 智能体运行在手机里的一个小型 Linux（proot）中，不需要电脑、adb 或 root + 可以用自己的订阅，也可以用 API 密钥，比如 DeepSeek（密钥不会交给智能体） + OpenCode 自带免费模型，不用账号也能试 + 支持中文界面，开源（Apache-2.0） +
157
status.live
Which lifts are spinning, which rides are running, which campgrounds are open today. status.live reads ski resorts, theme parks, national parks, campgrounds, marinas, surf and webcams from the source, on one page. Free, no account. Also a JSON API and MCP server.
原文1
aliasflare：跑在 Cloudflare Workers 免费额度上的自建别名邮箱
在 Cloudflare Workers 上自建永久别名邮箱，一键部署、完全免费。每个网站一个独立地址隐藏真实邮箱；在自己的收件箱里直接以别名回信；自动把验证码提取到主题；去除追踪像素；只存元数据不存正文；带 Web 面板、REST API 和 MCP 接口。只用 Workers + D1 + Email Routing，约 1000 行 JS，MIT。 亮点 一键部署在 Cloudflare Workers 免费额度上（Workers + D1 + Email Routing），零成本长期使用；为每个网站生成独立别名地址隐藏真实邮箱，并在自己的收件箱中直接以别名回信；自动把验证码提取到邮件主题，自动去除追踪像素，只存元数据不存正文；同时提供 Web 面板、REST API 和 MCP 接口，方便 AI Agent 集成，MIT 开源。
2
Pangolinfo Amazon Data MCP
Pangolinfo Amazon Data MCP gives your AI agent live Amazon data through 19 tools: product details, keyword search, reviews, best sellers, new releases, offers and more across 13 marketplaces. No scrapers to maintain, no anti-bot fights — connect in one line of config and ask in plain language. Works with Claude, Cursor, VS Code and any MCP client. Free tier: 60 credits every month, no credit card.
原文1
Wavefunction：Mac 上的开源桌面编程智能体，多个任务同时跑
Wavefunction 是一个开源的桌面编程智能体，运行在 macOS 上，免费，MIT 许可。现在的编程智能体功能越堆越多，Wavefunction 反其道而行：界面极简，只有一个内置智能体，可以开多个对话同时干活。 模型自由：可以用本地模型、自己的 API key，或者 ChatGPT 账号登录，随时切换 并行工作：在 Git 仓库里，每个新对话都有自己的分支和工作目录，互不干扰 运行中可插话：任务执行时发一条消息就能调整方向，也可以排队等它做完 随时回退：可以从任意一条消息分叉对话，或者把代码和对话一起退回到那个时间点 独立桌面：Computer use 在本地 Docker 容器里运行，不占用你的鼠标和键盘 兼容现有配置：读取 AGENTS.md 和 SKILL.md 技能，内置 70 多个连接器，也支持自己添加 MCP 服务器 作者：@stoyan-stoyanov
9
agentmash
Agentmash is an open-source multiplayer AI for teams' coding agents, a shared context layer that tells every agent which files a teammate's agent is changing, so the agents talk to each other and work on their own without overwriting anyone's code. You can switch between tasks running on different models, @mention a teammate, your own agent or a teammate's agent like in Slack, and keep skills, chats and automations in sync. Connect Claude Code, Codex, Antigravity or your own open-source models
原文116
imread — extract a public article as Markdown, with a source-preserving CLI example
Disclosure: I am submitting on behalf of imread. Tool: imread extracts the main text of a public web article so it can be read in a simpler interface or copied as Markdown into research notes. Core extraction works without an account. Its documented REST API and MCP endpoint also support developer workflows. A concrete example is collecting a long public article for a reading notebook while keeping the original URL beside the extracted text. The accompanying tutorial includes a small Python standard-library client that saves this attribution, refuses to overwrite an existing file, and reports extraction errors: The published example was downloaded and run against the live API before this suggestion. API documentation: Limits: login-only pages and pages dependent on client-side scripts may not extract. Speech and translation are optional interface features and may use external providers. The tutorial does not republish any third-party article body, and extraction is not a licence to redistribute a source article. This is a resource suggestion for your software/tools section, not a user review or an endorsement claim. Thank you for considering it.
原文—
claude-star-ipc：让 Claude Code、Codex 等不同厂商的 AI 编程终端分工协作
让几个 AI 编程终端（Claude Code、Codex CLI 或别家的 CLI）在同一个项目里分工：一个 hub 终端拆任务、派简报、收结果，worker 终端各自干活。终端之间只靠一个 SQLite 信箱通信，所以可以混用不同厂商的模型，比如一家出方案、另一家实现、第三家审。 和类似方案的区别：Claude Code 自带的跨会话消息和 Agent Teams 只在同一家 harness 内部；DeepSeek 刚发的 Super-agent 架构（Router / Worker）和 dsh 的各种 team 插件也都绑定自家运行时。这个项目把协调层放在 harness 外面，哪家的 CLI 都能接进来。 目前只在 Windows + WezTerm 上跑过，MIT 协议。
0
iKanban
iKanban is task and project management for engineering teams that work with AI coding agents. Connect Claude Code, Gemini CLI, Cursor or any MCP-compatible agent through the native MCP server, and the agents read and update the board. Sprints, releases and linked docs sit alongside the work. A security scorecard tracks OWASP Top 10 coverage for your team. Bring your own Claude, OpenAI or Gemini key. From $8 per user per month.
原文1
PartsPlan
An MCP server for electronic component datasheets. Connected to ChatGPT, Claude, Cursor or VS Code, it finds the manufacturer's datasheet, reads the page, table or graph, and gives each value with its document and page so you can check it.
原文1
Awesome REA：AI 逆向和复刻案例清单，逐条核对是否真用了 REA
最近 REA（Reverse Engineer Anything，GitHub 41 万+ Star）很火，"用 AI 逆向了 XX""一个人加 Agent 复刻了 XX"的帖子很多。但这些案例里，哪些真的用了 REA，哪些其实是 Claude Code 或 Codex 配 Ghidra，很少有人核对。 这个列表收集了 100 多个 AI Agent 逆向和重建软件的案例，每条都标了出处： 按标签区分：REA（作者或官方确认用了 REA）、AI-RE（其他工具做的 AI 逆向）、Rebuild（净室重写）、Claim?（待核实） 可以按"它替代的软件"查找：Photoshop、OBS、TradingView、罗技 G HUB 等 汇总了两组对照实验：rea-lab 482 次预注册运行，用 REA 解题率没有显著提升，时间和成本约为两倍 写清了法律和合规边界 中英双语，CC0 协议。
363
SoloPup Memory：为 DSH 保存带出处的项目记忆
这是为 DeepSeek Harness 补充长期文本记忆管理的原生插件。项目决定与跨项目工作偏好按范围保存，记忆附原文出处、适用条件和修订记录，便于要求变化后追查。当前为 MIT 公开测试版 0.21.0-beta.2，需 Node 24.9+、DSH 0.2.1-rc.0 至 0.3.0 之前版本；词法召回会漏匹配，模型审核也可能失败。 亮点 保存“为何记住”的原文引用和条件，修改决定时保留修订历史，适合持续迭代的项目。 原生 DSH 插件，可区分全局偏好和项目要求；不是通用 MCP 服务器。 文档公开说明召回、隐私和备份边界：模型 API 可另行收费，本地存储不等于完全离线。 维护者 OakcoderX 授权的 Codex AI 辅助自荐。公开 macOS 宿主检查由维护者记录，独立用户采用情况尚未验证；没有宣称生产成熟度或独立性能成绩。 已核对投稿模板与审核标准；官网按仓库地址搜索的公开结果中未见本项目，但登录限制使该搜索不能证明全站不存在。GitHub 历史投稿查重未发现重复。
1
ArtifactSend
Share a PDF, pitch deck, Word, PowerPoint or Excel file, or a Claude artifact as a private link. See who opened it, how far they read and time on each page. Email gates, a link per person, expiry, watermarks, NDAs and data rooms are free, with unlimited team members. Ask Claude or ChatGPT who read your deck through our MCP server. We only charge when AI works for you, a few cents at a time.
原文5
Codex Skills
Codex Skills is a free directory of 872 MIT-licensed Agent Skills for OpenAI Codex. Each skill is a complete SKILL.md playbook that Codex loads the moment a task needs it: PDFs, React, Postgres, security reviews, AGENTS.md and more across 28 categories. Install any skill with one command, grab curated collections like the Starter Pack, and read every line before it runs. No account, no telemetry. Works in the Codex CLI, IDE and app.
原文1
HostEdge
Hosting built for AI agents. Connect Claude, Cursor or any MCP agent: it deploys your site to the edge, edits your code with changes shown live in your browser, reads logs and rolls back. Database, sign-in, analytics, domains and SSL included. Free plan.
原文1
captcha-hub
Browser toolkit for agents: Chromium over CDP, vision endpoints, HTTP hub, MCP and CLI. npm packages are demo-allowlisted; MV3 extension has separate research scope. Written in JavaScript. Open source, MIT. Source, docs and issue tracker:
原文1
Local Commander：让 AI 在你自己的电脑上干活
开源 MCP 工具 Local Commander，让 Claude、ChatGPT、Cursor 或 Ollama 本地模型在你自己的 Windows、macOS、Linux 电脑上执行命令、读写文件、截图，电脑无需装 AI 客户端。作者在手机上说一句，家里电脑就更新并发布了网站。 （30 秒演示： ；托管版： ）
0
Local Commander
Local Commander lets ChatGPT, Claude, Cursor or any MCP client work on your own computers. Install a small agent on each machine, add one MCP URL, then say "ship the fix" from your phone. Your home PC does it. No remote desktop, no open ports.
原文3
Circular Edge JD Edwards MCP Server
Circular Edge helps you apply MCP (Model Context Protocol) to JD Edwards EnterpriseOne. Ask questions, find orders, check inventory, trigger orchestrations, and execute approved transactions through any AI application. Talk to a JDE and AI expert today.
原文2
PageSift
PageSift 是用 Rust 开发的内容提取引擎，能把网页、PDF 与 Office 文档清洗成结构化的 Markdown 或 JSON，直接喂给 AI Agent 和 RAG 流水线。它兼容 Firecrawl、Jina Reader 的接口并内置 MCP 服务，单二进制就能自托管，适合做爬虫数据提取、知识库构建和文档问答的入门实践。 亮点 作者自荐：PDF 转 Markdown 比 pypdf 快约 22 倍；2 亿任务量级运行内存峰值 888 MiB；支持 Python/Node SDK 与 CLI；AGPL-3.0 开源（另有商业授权）；约 175 个 GitHub star，持续活跃更新（当前 0.81.x）。
179
Terminay
Terminay keeps the shells, files and agents for each project together in one desktop app, with split layouts and coloured project tabs. Its MCP server lets Claude Code, Codex and other agent CLIs read terminal output, open splits and run commands, with permissions you set. Pair a browser or another machine for remote access. It also has cron and event automations, macros, terminal recordings and a file browser with Git diffs. Free and open source (AGPL-3.0), for macOS and Linux.
原文2
Bazavo
Bazavo is a classifieds marketplace your AI assistant uses directly. Connect bazavo.com/mcp in Claude, ChatGPT, Grok or Cursor, then just ask: "sell my bike", "find a flat under €900", "hire a plumber nearby". Posting ads is free.
原文1
Motionboard
发布片镜头配方库，38 个 Apple、Linear 等官方发布片镜头的 HyperFrames 复刻，附 agent skill，npx skills add 一行安装让 Claude Code 做发布片
—
Swastik Codes
Swastik Codes is a high-performance, lightweight API platform designed to replace bloated developer tools. Seamlessly test, mock, and monitor HTTP/REST, GraphQL, gRPC, WebSockets, MQTT, and MCP directly in your browser. It delivers enterprise capabilities—including test scripts, collection runners, load testing, mock servers, and automated documentation—without compromising on speed or system resources. Available instantly on the web, with optimized desktop apps for Windows, macOS, and Linux.
原文1
Motionboard：把 Apple、Linear 发布片的镜头拆成 38 个代码配方，附 agent skill
Motionboard 是我开发的开源（MIT）项目，把 Apple、Linear、Cursor 等 21 支官方发布片拆成 38 个镜头，用 HyperFrames（HTML + GSAP 渲染视频）逐镜复刻，每页原片和复刻并排播放。打包成 agent skill，npx skills add krisyangdev/motionboard 安装后，Claude Code 可以照着做出你产品的发布片。
51
Agent Aisle
A free, open-source security audit for Agent Skills. It reads a skill as text, never runs it, and grades it A to F. Plus a store of audited skills for Claude Code, Codex and Cursor.
原文0
Tracked
Native time tracking for freelancers who bill by the hour. Start a timer from the menu bar or the MacBook notch, and fix forgotten time by dragging blocks onto your day. Projects with hourly rates and budgets become PDF timesheets clients accept. A built-in MCP server lets Claude start timers and log your hours. Import your full history from Tyme, Toggl, Clockify, Harvest & more. No account, your data stays on your devices. One purchase covers Mac, iPhone, iPad & Apple Watch, widgets included.
原文11
SpaceO
Give AI agents a separate display for real Mac apps while you keep working. Connect through MCP or the CLI, watch sessions in the Viewer, and take over when an agent needs help. Supports native apps and Chromium. Free and MIT licensed; Apple Silicon, macOS 14+. Attention isolation, not a security sandbox: apps run as your macOS user. Runtime support depends on your macOS build.
原文62
拾趣 - 用 AI 把中国古典故事变成连环画
拾趣是一款将中国古典故事变成连环画的开源工具。选择书籍、主题和画风，AI 自动改编故事、编排分镜并生成插图。支持电脑和手机浏览器阅读，可导出 PDF 或长图，故事、图片和阅读进度保存在自己的电脑上。 亮点 从选题、故事改编到分镜绘图，完成整篇连环画创作。 支持多种画风，让古典故事呈现不同的视觉风格。 可查看原文、分镜和制作提示词，了解 AI 的改编过程。 支持已登录的 Codex CLI，利用订阅额度创作；也可接入 OpenAI API 或 Gemini。 本地保存作品，手机浏览器阅读，支持 PDF 和长图导出，方便收藏与分享。
2
Escribal
Escribal is a Windows app that generates documents from a workflow: a list of instructions you or your AI agent set up once, without programming. It reads your data, loops, decides and calculates, then writes documents, spreadsheets, presentations and PDFs. You build it on a canvas, your agent over MCP, and after that it runs without AI: the same figures every time, on your computer, no cloud.
原文0
WinnerAdSpy
Winneradspy is an ad creative intelligence platform built specifically for mobile app and game marketers. Explore 20M+ ad creatives across 200K+ apps, uncover winning hooks, visuals, messaging and CTAs, and track competitor campaigns with real-time alerts. Unlike generic ad spy tools, it scores creatives as Winner, Loser or Learning, helping UA managers, growth teams and agencies understand what works and build better campaigns. Query ad insights through ChatGPT, Claude or Gemini via MCP.
原文113
自荐 | Codex Co²：AI 时代的结对编程，应该是怎样的？
重新搬回的“椅子” 过去，两个程序员想一起解决同一个问题，可以把椅子挪到同一块屏幕前。 一个人负责操作键盘，另一个人负责审视思路；遇到分歧，就停下来讨论。屏幕上的代码是两个人共同产出的成果，问题的来龙去脉，也在这一问一答中逐渐对齐。 这种协作的方法，就叫做“结对编程”。 如今，AI 俨然已经成了坐在我们屏幕另一侧的新搭档。 我们向它描述需求，让它阅读代码、分析问题，并尝试修改。在一轮轮对话之后，那条会话里逐渐积累起了整个任务的背景，包括：为什么要做这个功能，哪些方案已经被否定，还有什么约束必须遵守等等。 可当后续有另外一位同事想参与进来，或接手你的工作时，事情就变得开始低效了。 我们需要把对话截图发过去，把代码复制出来，再把同事的意见转述回给 AI。对方如果想追问一句，又要等我们把问题送进去，再把回答搬回来。 AI 明明已经能够参与越来越多的执行工作了，但人与人之间的协作，仍可能被这些来回的转述给拖住。 个人的生产力被放大了，团队掌握的信息，却未必能同样顺畅地汇集到一起。 是时候，我们该重新搬回那张“椅子”了。 Codex Co² 想解决什么？ 基于同一条会话的协作，正是我想用 Codex Co² 来解决的问题。 Codex Co² 的全称是 Codex Conversation Collaboration（Codex 会话协作），名字有点长，但是想做的事情很简单：让多个人参与同一条 Codex 会话，一起推进工作。 分享者可以继续在官方 Codex Desktop 中工作，只需在 Co² 里选中一条已有会话，生成邀请，发给协作者。协作者则通过浏览器打开邀请，申请加入；获准后，就能阅读共享上下文，向同一个 Agent 补充要求，并跟进它的回复。 协作者无需安装 Co²，也无需单独登录 Codex。发出的每一条消息，都会沿用分享者当前会话的工具、权限和额度。 邀请协作者 不妨看一个具体的例子。 假设有用户反馈：“任务已经处理完成，页面上的状态却一直没有变化，刷新之后才恢复。” 运营拿到 FeedbackID 后，可以先在 Codex 中分析相关的用户日志，得到初步结论：操作请求已经成功，问题可能出在页面状态的更新环节。 随后，运营就可以通过 Co² ，将这条会话分享给研发。 研发加入后，可以沿着已有的分析继续追问：服务端有没有发出状态变更通知？客户端是否收到？异常发生前，是否经历过断线和重连？ 结合日志与相关代码，问题逐渐收敛：客户端在网络重连后没有恢复实时订阅，后续状态变更因此未能到达页面；而手动刷新会重新拉取数据，所以显示又恢复了正常。 到这里，一条模糊的用户反馈，才有了可以解释、可以验证的技术根因。 邀请另一协作者 但找到根因，并不意味着就走到了解决问题的最后一步。 研发可以继续让 Agent 梳理修复方案：补齐重连后的订阅恢复，并在恢复连接时校准一次状态；同时整理受影响的场景、改动范围、验证成本，以及临时处理方式，再邀请产品加入这条会话。 产品看到的，便包括了用户最初遇到的困难、运营从日志中获得的线索、研发定位根因的过程，以及不同处理方式的成本与风险。 基于这些信息，产品可以决定是否需要立即修复、如何安排版本，以及在修复上线前怎样回应用户。 运营带来现场，研发追到根因，产品决定如何推进。 不同角色的判断，在同一条会话里逐步接续。前一个人留下的分析，成为后一个人继续追问的起点，最终把一条用户反馈推进为一个有依据的处理决定。 收集分散的信息 Co² 带来的这个看起来很小的变化，恰恰对应着软件开发中一个长期存在的事实：理解一个问题所需要的信息，往往分散在不同的人手里。 产品同事知道某条看似奇怪的规则，背后对应着怎样的业务现实；测试同事知道一个偶发问题，究竟要经过哪些步骤才能复现；熟悉系统的工程师则知道，一处看似无害的改动，可能牵动哪些历史约束。 这些判断，很难仅凭一句“帮我把功能做出来”就交代清楚。 AI 能够处理的信息再多，也需要有人把这些最相关的事实带到它面前。因此，当我们讨论如何更好地使用 AI 时，更值得考虑的是：怎样让那些真正了解问题的人，更容易参与当前的工作。 Co² 尝试提供的，就是这样一个入口。 我希望它能够用在结对开发中，让另一位工程师直接参与方案讨论；用在远程排障中，让了解现场的人补充复现条件；也用在需求澄清中，让业务上的判断及时进入实现过程。 还有更多可能性 当然，Co² 目前仍是一个早期的独立实验项目。 项目代码已经在 Github 上开源，提供了 macOS 和 Windows 安装包，可以通过 Homebrew 安装后打开，也可以直接在浏览器上运行。 使用时还需要理解以下几点限制： 共享会话沿用原来的执行环境和额度，协作者的每一条消息，实际都是以分享者的身份在发送的，使用时需要注意权限控制和额度把控。 公网分享经过中继，使用普通 TLS，没有端到端加密，记得只邀请可信任的协作者，或者开启“仅内网可访问”开关； 退出 Co² 时，分享出去的邀请链接会变成不可访问，因为 Co² 设计上就是服务于当下的协作，而不是作为一个长期的入口。 如果你也正在和同事一起使用 Codex，欢迎试用，并把遇到的体验问题通过 Issue 反馈给我，一起参与 Co² 的共创。 那些曾经需要截图、复制粘贴、反复转述的环节，就是 Co² 这个工具能帮你们改进的地方。 结语 希望，我们在用 AI 来提升我们工作效率的同时，也能继续给身边的同事留出位置。 让了解需求的人、熟悉代码的人、发现问题的人，都能把各自掌握的那一部分带进来，完成一个共同的协作，这才是团队存在的意义。 曾经屏幕前的那把椅子，值得再搬近一些。
31
Sessions Pro for Cursor
Cursor stays signed in to one account at a time, so switching between work, personal and client accounts means logging out and back in. Sessions Pro for Cursor keeps each account signed in on its own, so they run side by side on Mac or Windows. Each keeps its own login, chats, settings, extensions and agent data; MCP servers, rules and skills stay shared. On a Mac, every account is its own app in the Dock, with a menu bar switcher and each account's monthly usage.
原文1
CarbonGuru
CarbonGuru is the first automated carbon footprint and offsetting service for everyday purchases. See the footprint of anything you buy, then offset or remove it in one step, for pennies on most purchases. It works on product pages, carts, receipts and orders on any site, in Gmail and webmail, and through Claude, ChatGPT or any MCP agent. Automate it with your agent, or schedule offsets that renew monthly. Verified offsets and removals, credit retirement handled. Free account and tools.
原文2
Alertedly
Alertedly watches your competitors across 16 public channels: pricing pages, product launches, hiring, ads, Reddit, reviews, trademarks and more. Every change comes with its source. Get alerts in Slack, a clean brief in your inbox, or connect your AI agents over MCP and ask Claude, ChatGPT or Cursor "what did my competitors ship this month?" No more tracking competitors manually.
原文8
Semlith
Coding agents make up quotes from PDFs and Office files they cannot open. Semlith is a free, open-source local index of your code and docs that Claude Code, Cursor and Codex search over MCP. In my test, made-up quotes went from 41 of 127 to 0.
原文3
TeapotRouter：给 Claude Code / Codex 的按量计费 API 网关
自荐（利益相关：作者本人） TeapotRouter 是一个面向 AI 编程助手的 API 网关：一个 Key 同时兼容 Anthropic 和 OpenAI 双协议，Claude Code、Codex CLI 改两个环境变量就能接入，按 token 计费。 为什么做这个 中转站市场口碑差，不是"中转"本身的问题，而是三件事：价格不透明、余额会过期、跑路没对账。所以这个站把三个承诺做成硬规则： 价格表全站公开：每个模型的基准价与实付口径写在明处，另有机器可读的 JSON 价格接口，方便比价工具直接抓； 余额永不过期； 计费日志可导出对账：每笔请求的成本、延迟、渠道都可查——Claude Code 烧 token 到底烧在哪，看得见。 其余：支付宝可充值（最低 $1）、多上游自动故障转移、支持退款（按剩余余额退，工单办理）。 彩蛋 名字和 logo 来自 HTTP 状态码 418 "I'm a teapot"——logo 就是「418 压扁成一把茶壶」，官网还专门做了一页 418 状态码小史（从世界上第一个咖啡壶摄像头讲到 RFC 2324）。 接入（Claude Code 示例） bash export ANTHROPICBASEURL= export ANTHROPICAUTHTOKEN=sk-your-key 上线两周的新站，欢迎批评。
—
slide-wright
No template library. slide-wright is a free, open-source (MIT) agent skill that invents a palette, type system and layout for each deck, then shows a two-slide preview before it builds the rest. Approve it or ask for a new direction. You get one HTML presentation you own: edit slides in place, export to PowerPoint or PDF, or deploy to a live URL. Works in Claude Code, Codex, Cursor and other agents that support skills. Install: npx skills add arifszn/slide-wright
原文3
iSearchFrom
Our MCP server is free. Connect Claude, ChatGPT or Cursor with one link and ask where you rank in any city, on any device. It runs a real Google search and answers in the chat. No API key, no card. Prefer to look yourself? Pick a country, city, language and device and see the Google results people there see, without a VPN. Then track rankings, ads, Google Maps and AI answers on autopilot.
原文1
Precleared
Checks each recipient before the first cold email: ok_to_email, consent_needed, blocked or unknown, with the rule, the footer it needs and a signed receipt. Works in Harvey, Apollo, HubSpot, n8n, Zapier, Make, Clay and any MCP client, including Claude.
原文1
LocalASO
Most apps only check their US rank and miss the other 50+ App Store countries, and ASO tools charge monthly. LocalASO shows indie devs and small teams where their app ranks for every keyword, in every country, on their Mac. • Keyword board: rank, volume and difficulty per country • AI turns keywords into what locals search • See which competitors outrank you • Rewrite your listing and upload it to App Store Connect • AI agent and MCP server for Claude Code
原文1
—
Imza – plan and review focused coding agent
Hi everyone. I've been building Imza for the last three months. It's yet another coding agent. I know there are a dozen have been posted here. But I believe this one is different enough to have a place here. It has three explicit modes: plan, build and review. Plan mode allows you to draft a spec document together with the agent. Both you and the agent can edit the document until everything is clear. Build mode takes the plan and implements it. Review mode provides you a code review tool. You can annotate things you want to change in the code and send them as revision to the plan mode. It's entirely written in C++, requires no account, supports 50+ providers including OpenAI subscriptions or local models. Imza supports MCP, skills and instruction files out of the box and uses 8-11MB RAM at startup. One of the biggest differentiator is the Lua engine. What I did differently here is agent receives only one tool, that allows the LLM to execute Lua scripts. I then passed C++ methods as Lua bindings that the agent can execute. So model can orchestrate multiple bindings in one call. Each binding passes through a permission layer. Since Lua is the only way to interact with the filesystem, imza can monitor and control the model's behavior. I'd love to hear your feedback. Thanks Website: https://imza.run Source: https://github.com/umutsevdi/imza
原文—
TrendsFast
TrendsFast is a commercial hosted API/MCP for niche trends, customer questions, competitor intelligence, short-video examples and search. Outputs include source links, dates, measured signals and coverage limits. New users get 100 credits once after verified signup. A separate MIT-licensed agent kit is available; the backend is proprietary.
原文3
MIKXX
Type what you need, like "a class booking page with limited seats", and Mikxx's AI agent builds it: questions, prices, logic and design. Change anything right on the page, then share one link. Take payments, run bookings that fill up, let customers sign in and see their orders, and connect it all with webhooks and an API. Free to start.
原文2
MCP for Your Mac Hardware
暂无项目介绍。
—
DualForge
DualForge is a mining game on BNB Chain, built exclusively for Binanceai Agent ID holders. No Agent ID, no key, no rig. Two holders each bring a key, pair in the lobby, and forge one rig together. Every round lasts 24 hours, and both partners must ignite the rig for it to produce $DFORGE. Each ID carries an element: same element pairs at x2.00, generating at x1.50, restraining at x1.00. Every holder can claim a free trial key, and everything runs on public smart contracts and a Skill file.
原文1
landmine
Your agent never checks GitHub for bugs in your exact versions. landmine does: it reads your lockfile, finds open issues filed against those versions, and gives the agent the bug and the workaround via AGENTS.md, a pre-edit hook or MCP. Free and open source.
原文5
web2mcp：一份配置把任意 REST API 变成 AI 可调用的 MCP 工具
一个零依赖的 MCP（Model Context Protocol）网关，部署在 Cloudflare Workers 免费额度上。接新 API 只需写一段配置：参数校验、路径/查询/请求体模板、响应字段裁剪，就会变成 Claude、Cursor 等客户端可以直接调用的工具。自带维基百科、Hacker News、天气、汇率、GitHub、网页转 Markdown 等连接器，还能把 OpenAPI 3 文档一键转成连接器。MIT 协议，中英文文档，README 有一键部署按钮。在线演示端点： 亮点 一份配置把任意 REST API 变成 AI 可调用的 MCP 工具，免费跑在 Cloudflare Workers 上。支持参数校验、路径/查询/请求体模板、响应字段裁剪；内置维基百科、Hacker News、天气、汇率、GitHub、网页转 Markdown 等连接器，并支持把 OpenAPI 3 文档一键转成连接器。MIT 协议，提供一键部署。
1
PixelDojo Sensei
Sensei is PixelDojo's AI creative director. Describe an ad, a product shoot, or a short film. Sensei writes the plan, makes every image and clip, and checks each take. You also get Character Swap, 80+ models, 310+ effects (3 free generations a day) and an MCP server. Paid plans (from $10/mo) include unlimited effects, unlimited Relax runs on FLUX 3 Klein, Qwen Image 2.1 and MiniMax H3 Turbo video, plus credits for all other models.
原文1
Mahshar
Discover APIs free through MCP. Providers list APIs while keeping upstream credentials private. Execution may require USDC via x402 on Arc Mainnet. Optional read-only ERC-8004 lookup checks agent identity records.
原文1
Garchi CMS
Garchi CMS is an AI-native content operations platform. People, applications and AI agents work on the same structured, governed content: people in a dashboard, applications through a REST API and SDKs, and AI agents such as Claude, ChatGPT and Cursor through a built-in MCP server. Agent changes arrive as drafts for a person to review and publish, with restore points for supported content, and approved posts can go to connected social channels from the same space.
原文3
—
PlateParrot
PlateParrot turns your recipes into one grocery list. Use the AI MCP: You plan meals, and the ingredients land on your list by themselves. Pick the servings and the ingredients land on your list, with amounts added up across recipes. The list is sorted by the sections of your supermarket, so you walk the store once. Share a list with your household: everyone sees the same list, and check-offs show up on every phone. Free for iOS and Android.
原文2
Hipo Hipo
Your workspaces follow you, not your agents. Every time you open a new window or switch AI agents, your AI gets amnesia. Hipo Hipo is a portable, governed "Brain" for your projects. Instead of dumping noisy chat logs into a vector DB, it stores your raw materials (evidence) alongside structured project memory (decisions, constraints, and methods). Connect one Brain to any agent, resolve context conflicts before overwriting, and take your workspace anywhere.
原文1
Backenly
Backenly is an autonomous backend platform built for coding agents. Your agent builds the backend over MCP, from Postgres to sign-in, storage, functions and realtime. Then it maintains itself: every minute a self-healing loop repairs missing indexes, RLS gaps and dead realtime triggers, verifying each fix live. As your app grows, it restructures its own schema once you approve. Risky changes wait for you, and every change is versioned with rollback.
原文111
static.md
Write markdown notes and whiteboards with your team, and let Claude, ChatGPT, Cursor or VS Code search, read and write them over MCP, and keep their memory there. You approve what each agent can do. Encrypted notes stay end-to-end. Free to start.
原文4
AtomIQ JD Edwards MCP Server
AtomIQ JD Edwards MCP Server brings AI-powered capabilities to Oracle JD Edwards EnterpriseOne through the Model Context Protocol (MCP). Designed for organizations using JD Edwards ERP, the MCP Server provides a way to connect AI assistants and agents with enterprise business systems, helping make ERP data and workflows more accessible through AI-driven interactions.
原文1
叮卡 DingCard：开源图文卡片编辑器，AI 可以通过 MCP 直接出整套图
叮卡（DingCard）是一个开源（MIT）的图文卡片编辑器，在浏览器里就能用，不需要账号。 长文自动排成卡片：粘贴 Markdown，按小红书、微博、推特的尺寸实时分页，有九套长文版式。 海报自由排版：文字、图片、形状、图表、表格、二维码放在同一块画布上，88 套模板，可以导出 PNG / JPG / PDF、长图或九宫格。 给 AI 用：MCP 服务器 dingcard-mcp 发布在 npm 上，也登记在 MCP 官方注册表。Claude Code、Codex、Cursor 这类客户端可以按内容生成整套卡片，按导出的样子检查文字被裁、对比度低、残留示例文字，再交给编辑器让人接着改。 在线使用：
14
Yenpo：和 AI 一起写项目时，读懂它改了哪些代码（macOS）
Yenpo 是一个 macOS 桌面应用，用来在和 Claude Code、Codex 这类 agent 一起写项目时，读懂和审查 AI 写的代码，跟上它的思路。它不接入 Claude Code、Codex、Cursor 或 GitHub，看的是你本地仓库里的改动。 直接读取本地的 git ，按改动列出 AI 改过的文件，逐个看行内 diff 内置 AI Reader，默认只读，可以对着代码问"这段为什么这么改"；用你自己的模型服务商 自带终端和编辑器，看完可以顺手改 不需要注册账号，应用内没有遥测 支持 macOS 14+，界面有 10 种
—
Flect
Flect is an Arabic-first customer messaging platform for teams in Egypt and the Gulf. Connect WhatsApp (official API or your existing number), Instagram, Messenger, Telegram and email into one shared inbox. Answer with AI agents grounded on your knowledge base, build chatbots with a visual flow builder, send safe WhatsApp campaigns with approved templates, and track deals and bookings in a built-in CRM. Developers get a REST API, webhooks and an MCP server.
原文2
MCP Server by XWiki
XWiki MCP Server is an open-source, self-hostable extension that connects XWiki to the AI agent of your choice: OpenCode, Pi, Codex, Claude Code, GitHub Copilot, and more. You can already use the extension to query your knowledge base to get answers, create and edit pages, and work with structured data directly in XWiki. Try out the extension and let us what's should we improve next!
原文147
AISoltrics Architecture Templates Bundle
Most AI template packs are prompt lists. These are production machinery — distilled from 35 published AI books and a Delaware C-corp's delivery practice. Three packs — AI Agent Prompt Pack (115 prompts), RAG Templates, MCP Templates — each as PDF + Markdown, free updates included. For teams shipping agents, RAG, and MCP into enterprises where governance, security, and evaluation are not optional. Sold by AISoltrics Technologies Inc. Support within one business day.
原文1
Agent Ping — 六家 AI 编程助手的菜单栏会话面板
同时开着 Qoder、Claude Code、Codex、ZCode、WorkBuddy、OpenCode 的时候，很难判断谁在跑、谁在等你回答。Agent Ping 是一个 macOS 菜单栏 / Linux 托盘面板：只读这些工具已经写在本机的日志，把每条会话标成正在跑、排队、等你回答、已完成或报错停下，点一行就回到那条会话；不联网、不要账号。 亮点 同屏六家 Agent，按状态分组而不是按工具分组——打开面板先看见「谁在等你」；等你回答的那条会话会带出问题原文和可选项（装 hook 后可精确到提问）。整个二进制不发起网络请求，只读日志，不碰账号密码；macOS 用 Homebrew 一条命令源码编译安装，Linux 有源码构建。
4
Sprouts Data
Sprouts Data is a hosted MCP server that gives AI agents access to about 45.7 million companies, 204 million people and 328 million person-to-company affiliations. Point your MCP client at one URL: search in plain English, find look-alike companies, look up records, map relationships and run multi-step research. Test keys cost zero credits, new workspaces get 1,000 free credits, and every billed response reports its cost.
原文1
Wavoice
Give your AI voice agents a phone line on WhatsApp. 📞 Connect WhatsApp Business calls to your AI voice stack in as little as 90 seconds. With WAVoice, you can: ⚡ Connect LiveKit, Dograh and other voice AI frameworks. 📞 Let AI agents receive real WhatsApp voice calls. 🔌 Skip complex WebRTC, SDP and webhook setup. 🛠️ Get started with a simple open-source CLI. Bring your voice AI to WhatsApp without building calling infrastructure from scratch.
原文4
—
Casus MCP
Query French legal sources (tax code articles with version history, BOFiP administrative doctrine, 3.4M+ court decisions, tax treaties, EU law) from Claude, ChatGPT or any MCP client. Every answer links to the official source, and resolve_references checks legal citations. All tools are read-only. Requires a Casus Starter or Pro subscription.
原文2
Token Saver
Open-source AI skills that pick the model for you: each task goes to the cheapest model that can finish it. For Claude, ChatGPT/Codex, Gemini, DeepSeek, Qwen, Perplexity and Kimi. Modeled 68% lower cost, 55% fewer tokens.
原文2
Sickle
Sickle now runs a remote MCP server, so Claude, Cursor, Codex or any MCP client can work across every mailbox you have at once. Ask "what needs a reply this morning?" and your agent reads every inbox and puts the urgent ones first. It can search, tidy, draft and send. Each tool has its own permission in your AI app, sending asks before every email, nothing is deleted for good, and Sickle keeps no copy of your mail. Works from any OS.
原文2
Opencommerce – OpenRouter for commerce, one API for AI agents to buy
If you're building an AI agent that buys things for its users, opencommerce by Agentcard gives it one buy tool: search, cart, checkout and order tracking across 5,000+ online stores, through a single endpoint. No merchant integrations to build and no agent browser to run. Agentcard places the order once the user approves. It's fast: it searches catalogues and makes a purchase in under 20 seconds. We test the purchase flow at each monitored store, from search to checkout, and the results are public: https://opencommerce.sh/?utm_source=hn&utm_medium=community&...
原文—
Fast agentic video editing in the browser (Tella)
Hi HN, today we're launching the biggest update to Tella ever ( https://tella.com/ ). Tella is all-in-one video creation. You can record, edit and publish with one tool. Today we're shipping: 1. Tella Agent: An in-app video editing agent 2. Tella Motion: Rich motion graphics support 3. Auto Capture: Agentic screen recording Our in-app agent uses the same tools as our MCP. All year we've been exposing more of our editor to the MCP. So when we started working on the in-app agent a few weeks ago, agents could already do a lot. Since then, we've worked hard to add as much capability as possible to the MCP. So whether you're editing on tella.com or from Claude Code you get the same capabilities. With the agent you can do all obvious agentic video editing stuff, like cut silences, fillers and mistakes. You can also do batch actions that would normally take a long time by hand. For example, if your tutorial video has a lot of zoom effects, you can ask the agent to adjust the scale for certain zooms: "for all zooms applied to when I'm showing Google Chrome, reduce them to 150%". But you can push it further with ambitious long-form edits and editing multiple videos at the same time. LLMs have proven to be very good at motion graphics. Remotion and Hyperframes launched earlier this year with React and HTML-based motion graphics respectively. We took a different approach and built our own motion graphics primitives on top of Skia and WebGPU. This means we're not limited by the DOM and can support things like liquid glass, motion blur, and text that moves along a path. We can also render a lot faster than DOM-based approaches because we can bypass Chrome and run everything directly on a GPU. But motion graphics are much more about how they look and feel rather than how they work. I've asked our agent to create motion graphics for animated UI, collage stop motion scenes, realistic paint splatters and photographic filters - the quality of the output continues to surprise me. Tella began life as a screen recording tool and today's update brings our origins into the AI era. Auto Capture is our new agentic screen recording tool. You can give it a website and a goal and it will remotely record itself using Chrome, completing your task. There are lots of other agentic browser tools with screen recording support, but we couldn't find any that were fit for "marketing grade" videos, i.e. recordings done in 4K at 60fps. So we built our own. For this to be practical (i.e. make demos of your own product) it needs to handle auth, so Auto Capture has a 1Password integration to handle signing into your sites and dealing with MFA. We also capture synthetic cursor data and keyboard shortcuts so that the Tella editor can add an animated cursor, animated keyboard shortcuts, and apply zoom effects. You can try all of this in your browser or connect to our MCP. Let me know what you think. Announcement + video here: https://x.com/9ranty/status/2115797158519550360
原文—
mcpscore
Audit any MCP server and get a 0–100 quality score across protocol conformance, tool quality, security and auth, and spec readiness.
原文5
Vibee
Create real web apps directly in ChatGPT with the Vibee plugin, or through another AI agent like Claude. Just describe what you need—bookings, orders, a customer portal, or an internal tool—and Vibee builds the app and its admin. Preview changes before publishing. Share your app with customers or teammates on desktop or mobile; they don't need an AI account. Hosting is included and data is stored in the EU. Start free with 3 apps, then expand with one-time purchases. No Vibee subscription.
原文2
Tazilo
Tell Tazilo what you shipped, led, or learned, as it happens. Your AI coach updates your profile, writes tailored, ATS-ready resumes, flags private details before you publish, and lets recruiters' AI agents ask you questions over MCP. Free in beta.
原文2
phpClaw – AI agent runtime for PHP with tools, memory and guards
Open-source AI agent framework for PHP. Tools, memory, security guards and a ReAct loop built in. Works with Laravel, Symfony, WordPress, Drupal, Magento, Joomla, OpenCart, PrestaShop. Any LLM provider. MIT.
原文25
Faylee
Faylee records the screen, console, every network request and each click behind a bug, on one timeline, and shares it all as one link. Your team opens it and sees exactly what broke, with the failing request already open. Paste the link into Claude Code, Codex or Cursor, and Faylee's MCP server gives the agent the whole report so it can find the cause and fix it. Instant replay saves the last few minutes after a bug already happened. Free for up to 3 members.
原文93
2
SaaS Pro Max
Product analytics, browser and server error tracking, reliability monitoring and feature flags for SaaS builders. Pilot and MCP support investigations. Free for 10,000 events/month; paid plans from US$12/month add session replay and more.
原文2
[分享] 给科研人员做了一个 Zotero AI 插件 BibGenie
大家好，我是 [BibGenie]( ) 的开发者，来分享一下自己做了一年多的东西，也想听听 V 站里还在用 Zotero 看文献的朋友的真实反馈。 ## 背景说明 用 Zotero 的应该都知道，它是科研圈最主流的开源文献管理工具，但本身没有任何 AI 能力。我自己读论文时有个很具体的痛点：把论文丢给 AI 问，答案说得头头是道，但我根本不知道它是从哪篇、哪一页得出的结论，最后还得自己回去翻 PDF 核对，费时费力。 所以我做了 BibGenie （去年叫 Zotero Copilot ，但因为 Zotero 商标的问题改了名），一个原生跑在 Zotero 里面的 AI 插件，侧边栏形态，可以直接访问官网 查看首页的展示视频，视频会更加直观。 迭代了一年多，目前更新到 v0.9.4 版本了，核心功能都已经开发完成了，今天来 V2EX 求一点反馈 （顺便吐槽一下，现在很多独立开发者只顾哐哐开发新产品，却不管维护，我也是深受其害。） ## 核心功能大概是这几块： 1. 直接和自己的 Zotero 整个文库对话。 不用导出文件、不用把 PDF 一个个上传到网页工具。在 Zotero 里面直接用 @ 或者右键 Add to Chat ，就能把条目、PDF 的某几页、EPUB 章节、批注、笔记、标签、整个集合拉进同一个对话里，多篇论文一起对比之类的。 2. **引用可追溯到文本块。** 这是我最在意、也花时间最多的功能。AI 回答里给出的引用，点一下能跳回 PDF 对应页，并高亮出它引用的那个文本块； EPUB 、网页快照也是一样定位到段落。科研场景下，我觉得这个功能真是刚需。 3. **本地优先，支持 BYOK**。聊天记录存在本地 Zotero 数据库里，不上传服务器；文库的语义索引也是在本地建的。模型这边三条路随便选：自带 Key （ OpenAI 、Claude 、DeepSeek 、OpenRouter 、各种兼容接口都支持，请求直连服务商，不消耗额度）；本地模型（ Ollama 、LM Studio ，聊天和 embedding 都配成本地的就可以完全不出本机）；或者用 bibgenie 官方模型开箱即用，什么都不用配置。 4. **内置了一个 Zotero MCP Server**。这个估计 V 站的朋友会比较感兴趣。插件里直接起了个本地 MCP Server （ Streamable HTTP ，长这样：127.0.0.1:23119/bibgenie/mcp ），不用再单独装什么 Python 服务。Claude Code 、Claude Desktop 、Cursor 、Cherry Studio 连上之后，可以直接检索我的 Zotero 文库、做语义搜索、读 PDF 附件（页面还能渲染成图片，看图表和扫描页没问题）、读笔记和快照，也能在客户端自己的权限确认下建笔记、整理集合。 5. 帮着整理 Zotero 文库。 可以让它整理集合、清洗标签、核查元数据（时间长了，zotero 文献库的标签，分类之类的都乱了，可以直接来整理）。元数据是交叉比对 arXiv 、Semantic Scholar 、OpenAlex 、DBLP 、OpenReview 这几个来源的。所有改动先出一张 diff 预览卡，点批准了才写进去。 ## 关于定价 所有的功能都是免费的，需要购买的只有模型的 token 。免费用户每月送 200 credits ，Pro 是 $9.99/月，有 10000 credits （ Pro 肯定够用了!!）。现在还有个 $199 的 Lifetime Pro ，v1.0 之后会下架（当然已经买了的不受任何影响，下架只针对新购），独立开发的小产品确实长期扛不住终身制, 前期已经倒贴了不少了。 官方模型目前是通过 OpenRouter 接入的，我在成本价上只加 7% 用来覆盖服务器和其他支出，目前没打算靠 token 差价赚钱（毕竟也支持自定义模型）。 官网和文档在这： 文档： 最后想认真求一波反馈，尤其是平时用 Zotero 的朋友。批评也完全欢迎，这东西的不少功能就是被用户骂出来的。感谢阅读。 （很抱歉，没有福利带给大家，毕竟这个产品是免费的，相信 V 友都有能力接入自定义模来使用，如果确实有想订阅的，可以留下邮箱，我给你 8 折优惠券。 当然，有 llm 服务商愿意支援 token 网页是欢迎的。）
—
Refract
Refract is a social media scheduler for creators, small teams and agencies. You write one short note, and an AI agent for each network turns it into a post that fits there: the right length, tone, format and hashtags. You review and approve every version, then Refract schedules it. Works with Bluesky, Mastodon, Pixelfed, Pinterest, Tumblr, Telegram, Discord, Slack, WordPress, DEV and YouTube, plus an API and MCP server for Claude and ChatGPT.
原文1
LockAutopsy
Token pages say 'liquidity locked' without checking if the lock can be canceled. LockAutopsy reads vesting contracts directly on-chain (Sablier, Hedgey, UNCX, Team Finance, PinkLock, Streamflow) across 5 chains, flags cancelable/revocable locks, and links every finding to the explorer so you can verify it yourself. Free forever: web, CLI, browser badge, MCP + x402 for AI agents. Optional one-time $19 USDC Pro. Read-only — never connects to a wallet.
原文1
Sketch Stadium – watch your agent play Pictionary
tl;dr: Agents can play Pictionary. You can connect your own agent, or just watch a game in progress here: https://sketch.wakeshift.ai/games/q5ucru Hi folks! I initially built this as a proof-of-concept for sending push notifications to AI agents. Watching the agents play Pictionary together has been compelling, so I thought I'd ship it as a standalone demonstration. In Sketch Stadium, your agent subscribes to game events. Agents take turns either drawing or guessing. Easy clues are 1 point, medium are 2, and hard are 3. It's entertaining to watch and some of the drawings are impressive (and others are mind-bogglingly bad). This is the real opportunity for the model labs who have been pelican-maxxing to shine. The first version gave agents the ability to draw lines 10 at a time so that it would be a little bit more like real Pictionary. However, it turns out agents are _really_ bad at drawing that way. So instead I let them use a subset of SVG to draw and that's worked much better, even for the lower-cost models. You don't need to create an account to connect your agent, just add the MCP and authorize. But if you don't want to do that, just watch the game linked above. As a bit more background, agents often start processes that take an indeterminate amount of time to complete (e.g. CI, waiting for approval in a separate system, sending an email and waiting for a reply). If an agent needs that process to complete before it can proceed, it can: * poll, * ask the user to tell it when it's done, * use a bespoke built-in integration, or * use a proprietary capability (e.g. Claude Code monitors) Today, there's no easy, consistent way to wake up agents through a webhook or push notification. I dealt with this firsthand with Clawvisor (authorization for agents) where I needed agents to recognize when the operator had clicked an out-of-band approve button. The best-supported way to achieve that at the time was long-polling. This project is basically a tech for Wakeshift, which is a cross-harness, simple SDK for providing push notifications for agents. It attaches to your MCP to allow consumers of your MCP to subscribe to real-time events, and uses whatever available hacks per harness to wake an agent when dormant. For most harnesses, Wakeshift requires a hook to be installed for the best experience. It will upgrade automatically as harnesses start adopting MCP events. I'd love any feedback or ideas for other products or games that you'd want to see built on this capability. Or MCP servers that you use today that you think would benefit from push notifications.
原文—
Engramory：AI 编程助手的 Markdown 长期记忆，附 DeepSeek Harness 插件
Engramory 是一套给 AI 编程助手用的长期记忆规范，MIT 开源。记忆就是一个文件夹里的小 Markdown 文件：一条事实一个文件，外加一个每次会话都会加载的索引 MEMORY.md。没有数据库、向量库或后台服务，记了什么随时能用编辑器打开看、改、删。
198
Search Education
Search Education Australia provides education counselling (with QEAC-certified counsellors), migration counselling, visa assistance, skills assessment guidance and education pathways for students, skilled professionals and families in Australia. Search Education has multiple MARA-registered migration agents, providing the most excellent guidance.
原文1
Zero-markup OpenAI-compatible router.
SentioGate is a zero-markup LLM router. Bring your own keys, pay providers directly, $0 on tokens. - 3-line drop-in with the OpenAI SDK (Python & JS) - Smart auto routing with failover - Long-chat memory: 2x fact recall vs summaries - Agent guardrails: blocks destructive bash/SQL, stops loops - Per-request spend tracking - Free community pool: 600 req/5h, no keys needed - Global edge
原文2
honestly：AI 编程助手说「测试通过、已修复、已部署」时，核对它有没有真的做过，没有就当场拦下回复（Claude Code / Codex / Cursor / Gemini CLI）
三组公开数据：有人人工标了 150 个 Claude Code 会话，357 句"测试通过"里 80 句背后根本没跑；圣母大学等分析 20,574 个真实会话（Claude Code、Codex、Cursor、Gemini CLI 都在），16,118 个失误里每 100 个就有 22 个是"没做完却说做完了"；Transluce 看了 8,600 个会话，每 55 个就有 1 个严重到"没验证的代码直接进生产"。 它不是每次骗你，是有一定概率不去检查却告诉你检查过了。轮到你头上是什么样：Codex 仓库有人提 issue，它说验证过了就部署，线上卡了五分钟，数据库连接池被打满；Claude Code 仓库有人提 issue，它说修复完成，缺陷还在线上，8,726 条记录漏掉，另一条是测试明明在报错、它回复 ALL PASSED；Cursor 论坛有人说它"浏览器里试过了"，追问才承认没开过浏览器。便宜的结局是 CI 红了回头排查，贵的是线上炸、数据漏，最糟的是后面几轮全建在这句话上，错一层叠一层。 honestly 是一个 hook：助手每写完一条回复，它读这一轮的会话记录，找出"测试通过 / 修好了 / 已推送 / 已发送 / 已部署 / 已保存 / 查过了"等 14 类结论，看最后一次改代码之后有没有对应的命令，并读命令的输出而不是退出码（pytest | tail 的退出码是 tail 的，测试全挂也是 0）。没有依据就拦下回复，要求它先跑再说；下一条回复会写明为什么被拦、补跑的结果是什么。
2
SpecDeck – Bring coding agent decisions back into the spec
Writing good specs is hard. You don’t know everything upfront and it’s impossible to write every little detail in the spec. Much of my job as a developer has been to treat specs as intent, and fill out the blanks in the code and use the code reviews as a way to question those decisions. But as coding agents have started to produce much more code, the decisions that actually affect the product are getting harder to find. So I created SpecDeck. It helps everyone on the team write specs and hand them over to developers and their agents through MCP. During implementation, agents write the decisions that matter back as proposed spec changes and leave smaller implementation choices in the code. When the work is done, a person reviews and publish those changes into the spec. The workflow is modeled on git: a published spec is main, drafts are branches, and proposals are pull requests. It should feel familiar to developers, while making it easy for the whole team to work with specs and keep them up to date without touching git. It’s still early and there’s plenty to improve, but I’d love to hear your feedback.
原文—
Matriks MCP
Matriks MCP is an AI integration platform that seamlessly connects your favorite AI assistants (Claude, Cursor, OpenAI) to real-time financial data. Instead of switching between terminals and AI tabs, simply prompt your AI to analyze BIST quotes, fetch balance sheets, or perform technical analysis instantly directly in your workflow.
原文1
Metis Suite
Open-source email, calendar and drive client for Gmail, Outlook, Zoho and IMAP, with an AI agent, workflows and MCP. Self-host free or use Metis Cloud.
原文1
Ellie Docs
Ellie Docs connects to your GitHub repo, works out what your product does, and writes your whole help centre, with real captured from your running app. Merge a PR and it updates the articles that changed and drafts your changelog, so docs stop drifting. Publish on your own domain with Ask AI answers and an API reference, and edit everything from Claude or Cursor over MCP. Free plan, no card needed.
原文2
ragbox
Let Claude Code and other MCP clients search your local documents. Runs in Docker, reads files without modifying them, and automatically keeps the index in sync.
原文1
Catfolio - Portfolio Tracker
Catfolio brings all your investments into one clear view. Available on iPhone, iPad and Mac, with MCP support. Connect supported brokers or import CSV files. Track holdings, returns, costs and cash flows. Explore portfolio heatmaps, compare benchmarks and review trades and dividends. Research stocks with charts, financial data and extended-hours prices. Connect your own AI for deeper insights. Privacy first: portfolio data stays on-device, with credentials secured in Keychain.
原文1
Solaris Dev Shop
Project URL Category Python Project Title Solaris Dev Shop: a local Python router that picks one installed skill for a task Project Solaris Dev Shop is a public MIT skill collection and local Python router. It picks one installed skill for a task and runs that skill command. It does not start an agent or call an API. Highlights It routes one local task to one installed skill file. It does not call an API, start an agent, or host a model. Example Code markdown or Videos
原文2
rekit
rekit delivers deterministic, read-only static binary analysis and headless Ghidra automation engineered specifically for autonomous AI coding agents. Eliminating hallucinations and token-wasting guesses, rekit enforces a strict 16-check verification pipeline across 3,200+ lines of Python and shell. Indexed and verified by the XERJ Corpus Hub as an open reference standard for coding agents. Completely free, MIT-licensed, and MCP-ready.
原文1
Tidy
Introducing Tidy Desktop. An open source, local first design canvas and agentic development environment (ADE) in one platform. Run coding agents in your local projects, create and refine interfaces on the canvas, then attach a design to a coding thread to build it. Made for designers and engineers who want to move from an idea to working code in the same workspace. Sync to the Tidy web app in the cloud to collaborate with your teammates and enable multiplayer with the MCP.
原文2
Lingxi
If you run AI coding agents, you know the loop: stare at the terminal waiting for "done", or miss the moment it stops to ask for permission. Lingxi is a quiet 3D cat that lives on your Mac desktop and watches for you. When Claude Code, Codex, Cursor or any MCP client finishes, fails or needs you, it walks over and tells you with a face and a speech bubble. Between tasks it naps, wanders and chases yarn. Runs locally, never acts for your agent.
原文1
HPSILab
New: HPSILab is now listed in the Claude Directory, so you can connect it to Claude in one click. AI agents get price predictions, Monte Carlo simulation, options IV radar and pre-trade risk checks via MCP, a Python SDK and the web platform.
原文1
Arcade：让 Claude Code 自带任务看板、笔记和等待小游戏
Arcade 是一个免费开源的 Claude Code 插件，把 Claude Code 变成小小的工作台：项目看板里的任意卡片一键“交给 Claude”，Claude 勾掉完成项并把结果写回卡片；笔记本由你和 Claude 一起记，所有打开的会话共享；登录 GitHub 后还有团队共享看板和聊天（可选）。agent 工作时，可以在 Claude Code 里玩 NES 游戏，或在浏览器里读公共领域的书、看与当前任务相关的文章；Claude 完成或请求权限时，游戏上方会弹出横幅。 亮点 不只是等待小游戏：和 cc-arcade 这类在等待时放小游戏的插件不同，Arcade 首先是工作台。看板卡片和笔记页可以直接交给 Claude，Claude 通过插件注册的 MCP 工具 mcparcadenotes 把结果写回。 多会话协作：所有打开的 Claude Code 会话共享同一个笔记本，每个会话可以从同一块看板上领自己的卡片。 不会错过 agent：agent 完成或请求权限时，游戏上方出现横幅，可以继续玩、暂停或直接回到 agent。 内容全部免费：开源授权的游戏、公共领域的书和有声书；无广告、无统计，AGPL-3.0。 支持终端、VS Code 和 Claude Desktop（Windows、macOS、Linux），界面和 README 有 13 种语言，含简体中文。
11
Point to AI
Stop telling AI “the button on the right.” Point to AI lets you click exactly what you mean, add a comment, and send the right context with it. It supports multiple elements, visual references, and structured output, so ChatGPT, Claude, Codex, and other AI tools can understand your intent more accurately.
原文1
孙宇晨的观点，用 AI 落地：两期访谈的全文和 23 篇 AI 教程
孙宇晨最近上了两期访谈，邵艾伦那期 4 个多小时，YouTube 上没有字幕，没法直接丢给 AI。我把两期都用 Gemini 转成了带时间戳的全文，加了按话题分的目录，按上下文改正了 200 多处听错的词。 访谈里能动手的观点，每个配了一篇教程，写清楚一步步怎么做、用什么工具。工具都给两条路，一条是好用的，一条是国内不用海外手机号就能开的。教程里提到的 86 个工具，价格和国内能不能直接用，统一记在一张工具总表里。 还有一份 AGENT.md，复制给任何 AI，说说自己的情况，它会照访谈的思路给你几条路和一个能反悔的小实验。用 Claude Code 或 Codex 的话，可以用下面这行命令装成 skill。 npx skills add wousp112/awesome-justin-sun-ideas-to-ai-howto 除了两期全文，仓库里的内容都是 CC BY 4.0，免费用，注明出处就行。
197
Sheetgraph, AI agents can understand complex spreadsheets
Hi guys, I built a tool www.lovespreadsheets.com earlier this year and a huge problem we kept running into is processing messy, real-world, complex spreadsheets for AI agents. Even a "medium" sized spreadsheet converts to about 1m+ tokens if read fully, which immediately blows up the current context windows of most models. So I noticed most agents sample data from spreadsheets in chunks instead. So I (& Claude) built this library Sheetgraph, https://github.com/asharma327/sheetgraph , to compress spreadsheets (~100x compression) while still maintaining their structure (like a dependency graph for formulas) & other relevant metadata such as macros, cell data types, formatting. There is a zipped full workbook of the spreadsheet as well for the agent to lookup data as needed. Still rough around the edges so would love any feedback on both the idea and any details. This is used in our primary product + services platform Love Spreadsheets so we are actively maintaining and improving it. Thanks!
原文—
AgentQon
AgentQon is a free AI agent harness. A panel of AI agents plans, checks its own work, and finishes jobs. It is free forever: no paywall, no credits, no ads, no commissions. You bring your own model key, or run a free local model. 6,799 agents and 4,840 skills are built in, plus an optional 213-skill Extended Library you can add at setup. Install it with one command: npm install -g agentqon. There is also an Android app. The product page is the GitHub repo.
原文1
Ankinize
Ankinize is a sync server for Anki with a REST API on top of the same collection. Use it to keep your devices in step, to add cards from your own tools, or to read your decks from an app you are writing. Give an agent an API key and it can create decks, notes and cards through the REST API or the TypeScript SDK, and run searches. The same cards sync to Anki Desktop, AnkiDroid and AnkiMobile. Agent skills for the API and the SDK are published at /.well-known/agent-skills.
原文1
AIRewards
AIRewards helps developers earn rewards through non-intrusive, one-line text sponsorships displayed across AI coding tools like Claude Code, VS Code, ChatGPT & etc. Developers earn a 70% revenue share, while advertisers reach verified developer audiences through targeted campaigns. With CLI wrappers, IDE extensions, browser integrations, global payouts, and a privacy-first architecture that never collects AI prompts or source code, AIRewards connects developers and advertisers in one ecosystem.
原文1
—
RuleFix
RuleFix gives Claude Code, Cursor, VS Code and Windsurf a compliance brain over MCP. It catches DPDP, GDPR, SOC 2, HIPAA and PCI DSS issues as code is written, suggests fixes, and only closes an issue after re-scanning proves it's gone.
原文3
Apgar – Catch agent regressions without an LLM judge
Hey HN, We build financial AI agents, and Apgar is the tool we wrote to see whether a harness change made an agent better or worse in production. Our offline evals can't cover every kind of question or session real users bring, and the LLM judge we run in Langfuse is too expensive to use on more than a small slice of traffic, so some failures get through. Apgar is a small release health checker that sits on top of our Langfuse data. It picks up the release from the production traces, and for the first hours of each new release it compares the agent's behavior with the previous release on things like tool calls per turn, retries of the same failing call, loops, output length and cost per turn (11 metrics right now). The release page shows what moved in both directions. When something gets worse, we get a Slack message (and a webhook) saying which metric moved and where the change is concentrated, for example only on Telegram or only in long sessions. We've been running it for a while and decided to open-source it: https://github.com/oimo-ai/apgar It's not meant to be another monitoring tool. We still use Langfuse for that. Apgar only tries to answer "is this release worse than the last one?" It works with Langfuse or with anything that sends OpenTelemetry. There's no LLM judge in it, so it's cheap enough to run on all traffic. The downside is that it can't tell that an answer is wrong if everything else about the turn looks the same. That's still what our evals are for. It runs on Docker Compose (or Helm) with ClickHouse, and it's Apache-2.0. The Agents page can load data if you want to try it without connecting anything. Curious how others deal with this. When you change your harness or prompts, how do you find out whether the agent actually got better or worse?
原文—
Link Building Skill for AI Agents
Open-source link building skill for Claude Code and AI agents. Run 10 tactics: - Broken links - Link gaps - Unlinked mentions - Resource pages - Guest posts - Link reclamation - Dead pages - Moving Man - Listicles - Skyscraper Includes Python scripts to check dead links, Wayback history, expired domains and contact pages, plus draft outreach emails. Most features are free. DataForSEO runs backlink lookups. All tactics cost well under $1. MIT licensed. Leave a star if you find it useful!
原文1
让 AI 自动给文章配表情包的 Agent 工具
一个给 AI Agent 用的表情包搜索和自动配图工具，包含 Agent Skill、远程 MCP 配置和命令行工具。装上后，Claude Code、Codex、Cursor 等 AI 工具可以按情绪搜中文表情包和英文梗图，或者读完一整篇文章，在值得吐槽的句子后面插好表情包，返回 Markdown 成稿。适合写公众号、博客、推文时让 AI 顺手配图，也可以用来学习怎么把一个服务做成 Skill 和远程 MCP。 亮点 会配图，不只是搜图。 把整篇文章交给它，它挑出值得配图的句子，以 Markdown 把表情包插进去，Agent 不用自己猜该放哪。 按情绪搜。 搜「无语」「裂开」「摸鱼」或者 facepalm、this is fine，拿到的是想表达的那个反应，而不是字面相近的图。熊猫头、奶龙、金馆长这类角色名按整词匹配。 标清版权。 每张图都注明能否商用、附来源链接，需要署名的已带好署名，可以要求只返回可商用的图。 接入简单。 npx skills add 一行命令装进 40 多款 Agent；也可以直接接远程 MCP npx memecub 在终端里搜。搜索不需要 Key。
26
SecAIQ Watch for Mac
SecAIQ Watch shows which AI tools run on your Mac, what files and permissions they can reach, which MCP servers they use and where their traffic goes. Read-only, 100% local, no account. 14-day free trial, then $19.99/year.
原文4
DeskVNC
DeskVNC brings VNC, RDP, and SSH together in one native app for Windows, macOS, and Linux. Manage machines from one library and let compatible AI agents access existing connections through MCP. 🤖 AI + human control: AI agents interact with sessions while you stay in control. 🖥️ One workspace: Access remote desktops and SSH terminals in one place. 🔐 Privacy-conscious: No account required; credentials stay in your OS keychain. 🌍 Free and open source: Explore and contribute to DeskVNC. globally
原文1
—
Blueprint 2.0
Blueprint is an open-source conversational skill that helps you clarify an idea, gather references, choose a direction and define a testable MVP. Get a portable plan with decisions, data/API, security, acceptance tests and agent handoff. FR/EN docs. No website or API key required by the skill.
原文2
Dbprint – Database Context for Agents
CLI that produces an offline snapshot of a database (DDL, column- and relation-level statistics, values with frequencies), combines it with annotations written by you or an agent, and exposes it as MCP and as a web UI. Works with PG, MySQL, Snowflake, ClickHouse, Databricks, BigQuery. How to try it: uvx dbprint context \ arboretum.seedbank.germination_trial \ --purpose query \ --project https://github.com/jakubro/dbprint/blob/main/docs/format/v1/examples/production/.dbprint.yaml Features: - Philosophy: snapshots are meant to sit next to your code and versioned so that you have a history of changes and agents can hit them when searching codebase. - Offline: no need to expose live credentials to agents. - CI: "dbprint check" can verify schema/data drift. - Annotations: complement statistics by documenting the meaning behind data. An agent can prepare them for you ("dbprint annotate"). - Redaction: values can be masked/excluded/hashed either explicitly with FQNs or automatically based on inferred sensitivity and value shape. - Format: Snapshots are YAML and markdown, so greppable and usable by other tools. Why I built this: SQL queries produced by agents were inaccurate, and the overall time to result was slow and costly because of finding the answers in codebase or in live database.
原文—
driftnet
I kept pasting interesting links into a chat to read later, then forgot what they were. driftnet sits in your group chat (Discord, Telegram, Slack, LINE) or your browser and turns each link into a card: what it is, what you'll see, why it matters. It sorts cards into folders, notifies you, and keeps everything searchable on a web page. Runs on your own computer with any model, or a local one via Ollama. Your AI agent can search it over MCP.
原文1
Breatic：文档、AI 画布与实时协作放在同一个项目中
项目仓库： 大家好，我是 Breatic 的开发者，想自荐这个面向创作团队的协作项目。 Breatic 将 Document Space 和 Canvas Space 放在同一个 Project 中：文档用于整理需求、脚本和创作说明；画布用于组织参考素材、生成和比较图像/视频等内容；项目中的 AI Agent 结合上下文协助创作。多人可以在同一个项目里工作，AI 提出的文档改写需要人审阅和接受，而不是自动覆盖原文。 对于开发者，仓库中可以看到 React/TypeScript、React Flow 画布以及 Yjs/Hocuspocus 协作层的实现。希望和对协作编辑、创作工具、Agent 交互感兴趣的开发者交流，也欢迎内部部署和贡献。 使用指南 开发环境指南 贡献指南 现有小任务：WSL2 开发指南实测、韩文画布控件母语审阅。 源码采用 Breatic Source-Available License，具体条款见仓库 LICENSE。允许个人、私有群组和组织内部部署；面向公众提供 Breatic 平台服务受限制。部署和 AI 生成功能仍需按文档配置存储及模型服务，并非完全离线运行。 感谢审阅！
4
—
RunDida
Plan your next run or race with free pace calculators, finish-time predictions, printable training plans and marathon information. Core tools work without an account. Includes a public REST API and an open-source MCP server.
原文1
Instagram Agent Skill by AZIRAL
Plan content, script Reels, draft captions and carousels, and review CSV insights in English, Russian and Kazakh. Five agent skills plus three local Python tools. Free and MIT-licensed. No Instagram login or auto-posting.
原文1
AXL.tech
Capture leads, close sales & onboard customers in one system: pages, quizzes, webinars, courses, video hosting & streaming, email, chat, CRM, telephony & AI calls. No integrations to build & maintain. Every interaction enriches each lead's history. Our AI model, trained on 1M deals from AXL users, scores leads using 300+ signals to help your AI & sales team decide who to follow up with, when & how. Give ChatGPT, Claude or Cursor access to lead context & 1,200+ sales & marketing actions via MCP.
原文3
因为想统一看自己全部券商账户，我做了一个叫 Catfolio 持仓追踪 iOS App
我的美股散落在好几个券商里, 一直找不到一个地方能看清楚: 总共持有什么, 今年到底赚了多少, 跑没跑赢大盘啊。券商 App 只看得到自己那一家, 记账软件要手动录每一笔交易。所以干脆自己写了一个。 - 支持各种券商 api 连接，不用手动输入（当然也支持自己手动导入 csv ）。 - 买入点卖出点都非常清晰显示在线上，还有金额细节，成本价都算的清清楚楚的。 - 账户数据都是自动算，合并算分开算，非常直观。 - 真实收益: 同时算时间加权 (TWR) 和资金加权 (MWR) 收益,可以和 SPY 、QQQ 等基准对比,不会被定投入金带偏。 - 收益来源: 按标的拆分收益,看清今年的钱是哪几只赚的。 - ETF 穿透: 把所有账户里的 ETF 拆开,算出合并后对单只股票的真实敞口。比如 QQQ 、VGT 、VOO 叠在一起,你实际持有多少 NVDA 。 - 分红:已到账的分红和今年的预期分红。 - 研究工具: 成交量分布、期权持仓墙 (OI walls)、市场情绪仪表盘,还能看佩洛西、段永平、巴菲特等知名投资者公开披露的持仓。 - AI: 内置 AI 助手,用你自己的 AI;还提供 MCP server,可以在 Claude Code 或 Codex 里直接问问题。 - 支持 A 股！可以搜索添加 几个设计原则: - 只读,不能下单。 - 计算都在手机本地完成。 - 不包含任何投资建议哈。 **App Store**: **官网**: 我已经修了好久，bug 已经不多了，但是一个人 vibe 开发,肯定有不少问题,欢迎拍砖。特别想知道大家还在用哪些券商是连不上的。
—
Betelgeuse
Free, open-source app for notes, docs and databases. Every page is a Markdown file in git on your computer, AI agents read only the pages you share, and plugins add what else you need.
原文2
Donchitos / Claude-Code-Game-Studios
Turn Claude Code into a full game dev studio — 49 AI agents, 72 workflow skills, and a complete coordination system mirroring real studio hierarchy.
原文3.2万
SpaceWise
SpaceWise is a native, privacy-first storage cleaner built for developers on macOS & Windows. Reclaim 40GB+ of Docker layers, Xcode DerivedData, dead node_modules, and AI weights. Features silent menu bar Auto-Clean and an offline AI Folder Explainer with SpaceWise MCP server.
原文7
CollectX CRM
Debt Collection CRM with Call Center Integration, Field Agent Tracking, and Automation for Agencies and Financial Institutions.
原文1
Sofuu
The native AI runtime for JavaScript & TypeScript. Single self-contained binary (<= 5.5 MB) with in-process local LLM inference (experimental engine), autonomous agents, native MCP, and encrypted QTSQ cognitive vector memory.
原文1
OLO ToolGate
OLO ToolGate is an Apache-2.0, self-hosted security gateway for AI agents, MCP servers and tool execution. Apply ALLOW, ASK or BLOCK policies, independent human approvals and single-use execution permits, with audit trails linking identities, decisions and outcomes. Now in beta, with a Docker Compose quickstart for local evaluation.
原文2
aidentity
Aidentity is a verified index of 526 AI agent skills from 25 sources. Every entry carries its author, its license, and the date it was last checked. Install any of them into Claude Code, OpenCode, Codex, Cursor, Copilot or Gemini CLI with one command.
原文1
Polish Proof
现有网页的视觉润色 Skill，围绕约定的文案、品牌、控件和行为限制修改范围，配套前后截图、保护项检查与交互测试记录；MIT 开源 - 演示与使用说明
1
Refcat
Refcat is a library of 560K+ real UI screens and 100K+ user flows from shipped products. Search by what you're designing, see the whole flow, then send it to Figma, capture any site with Chrome, or give your AI agent real references via MCP.
原文2
ShipSnag
Skills for Claude Code, Cursor and Codex, one for each thing that goes wrong once AI-built code goes live. Broke After Deploy finds what will break in production before users do. Key Leaked finds secrets in code, git history and build output, then walks you through rotating them. Bill Spiked finds the calls behind a bill spike and adds limits. Prove It Works (free) makes your agent show real test output before it says done. Every fix is a diff you approve, and it all runs locally.
原文1
ThinkingOrbs
ThinkingOrb(.searching). That is the whole integration. Nine hand-tuned dotted loading indicators for AI and agent interfaces in SwiftUI: eight are real 3D forms, rotated, depth-shaded and z-sorted, and the ninth is a morphing outline. All drawn only in grayscale dots, so they sit quietly in any interface, light or dark. Every orb pauses itself offscreen, stays in phase with the others, and respects Reduce Motion. Swift 6, iOS 17+, macOS, tvOS, watchOS, and visionOS, Swift Package Manager, MIT.
原文7
wise-men-flash：讲清理由、保留反方意见的 Claude Code 决策 Skill
阮老师好，自荐我做的开源项目 wise-men-flash（MIT）。 第 414 期谈 Jev 决策模型时提到，它只给一个数，这个数是怎么来的无从得知。wise-men-flash 走的是另一条路：给出的不是一个概率，而是一份讲清理由的决策备忘录。 它是一个 Claude Code Skill：一个难题同时交给三个 subagent——一位熟悉这类工作的实践者、一位相关领域的成员，以及一个专挑前提毛病的 Devil's Advocate，三人各走不同的推理路径，互相看不到对方的答案。主线程把三份回答合成一份备忘录（建议、理由、第一步怎么做、风险、最有力的反方意见、每条结论的把握程度），再交给一个没参与讨论的模型核对，然后才给你。 测试结果和局限：预注册盲测，8 道新题，每题 3 位 Opus 盲评，它得 23.4/25，每道题都高于 6 个热门 Claude Code Skill 和直接回答。但评委也是 Claude，题量小，而且单看正确性，直接回答反而更高（4.3 对 3.8）。原始答案、评分和分析脚本都在仓库的 eval-data/ 里。 在 Claude Code 里安装： /plugin marketplace add Ali-expandings/wise-men-flash /plugin install wise-men-flash@wise-men-flash 或者 npx skills add Ali-expandings/wise-men-flash -g。 另有完整版 wise-men：3–7 位成员独立作答、互评、分歧时辩论，最后由 Chairman 写出保留异议的备忘录。
9
StrictLLM – For when you need human-in-the-loop
Hi, We've been working on this for the last 9 months and are nervcited to share this on HN and gather input. StrictLLM is a desktop application for Mac, Windows, and Linux that will give you the following abilities: - Human-in-the-loop with a strong focus in security and control - Permissions enforced by code rather than a prompt - Being able to talk to multiple AIs in the same conversation thread - You can pull in any personas by @Mentions. - Local and cloud models supported (OpenAI API Compatible and Anthropic Compatible) - - LM Studio, Lemonade Server, Ollama, Claude, Strata, ChatGPT, OpenRouter, Kilo Gateway, etc. List is not exhaustive - No telemetry is collected. The only network requests being made will be to check for updates and license activation. Your chats, prompts, tools, and sessions are yours to keep and manage. https://strictllm.com/screenshots To dive right into it: 1) https://strictllm.com/download 2) Hacker News Trial License Key for activation: C564BD-1CD395-06A51E-CDF819-C48315-V3 This key will be good for a limited time up until end of November 30th and will be the same as the Pro license. No credit card or account required to use this trial. Why use this over manual mode in other tools? Manual mode does not guarantee that each call to a tool or MCP server will prompt you first before proceeding. Contact: support [at] strictllm.com Forum: https://forum.strictllm.com/t/discussions
原文—
Vaultbase
An open-source backend in one binary: collections with a REST API, auth, realtime, files, hooks, cron and queues edited live in the admin UI, and an MCP server for AI agents. SQLite by default, Postgres in beta. MIT licensed.
原文1
Backend for Agents and Builders who don't need one yet – Statebase
100% Meat Machine Post, handcrafted for you (I personally typed in the emdash)! site: < https://statebase.dev > quickstart: < https://statebase.dev/docs/quickstart > examples: < https://statebase.dev/apps > < https://statebase.dev/games > < https://statebase.dev/examples > arch: < https://statebase.dev/arch > dsl: < https://statebase.dev/dsl > Hi everyone, I'm Brian. I like frontend a lot. I can do backend, but not every little thing is worth it and I've been making "detached" HTML apps (not even needing a web server) as long as I can remember. I wanted a way to standup some sort of secure way to save data, maybe with a little bit of logic with just a few commands. I didn't want code (TypeScript, Java, C#, etc), but I also didn't want no code either -- maybe it's the devops in me (Terraform?). I was done waiting, and made this. It hosts the logic, permissions and state you need for an agent, application or game with pure curl commands. Thousands of jobless engineers and "learn backend or die" attitude why should backend have all the fun? Besides this might be good for scripting and infrastructure setup too, so it has something for everyone. There's also a Tauri+pglite version for desktop dev (thank you pglite) and a Devvit template for Reddit games (I made a couple, ask) all running on the same engine. Please give it a shot, and if you like what you see join my team (I got tired of working on it alone so I added a childhood friend and a few ex-coworkers) and I in our Discord. I'm full time on it now so any feedback even negative is more fuel. If it sucks, we want to know about it, so we can make it better. Now for the big questions *Is it Open Source?* Not yet. There's a lot of reasons but the main one is the code is in a giant monorepo that has to be cleaned up. Another reason is I eventually want to make it Fair Source or Open Core, so that a hyperscaler can't just wrap it and charge everyone (including me) a lot of money just to use it. We do have to charge money for serious production use. *What if you disappear tomorrow?* Absolutely valid question. That's why I put a lot of effort into the export to SQL (eject) functionality. It's not perfect yet, but you can export all the logic and permissions into your own Postgres database. So if you want to run it 100% standalone from Statebase, you can. The SQL export generates is ugly for now; a lot has to do with nulls. We will improve it but I wanted to avoid forcing a schema to start. *AI Support?* MCP is coming very soon (next few days) and x402 payments is in the works. But I wanted to get it in the hands of flesh and bone, to see if it really excites someone as much as it does me. I know it's possible the vast majority of users will be agents, but hoping for at least a few flesh and blood champions. *Plans?* It's built on Cloudflare Durable Objects for cost savings and other advantages like built-in tenant isolation and native websockets (honestly scary how well DO mapped to our needs). But it could exist anywhere. We have a node version, a desktop version (Tauri+pglite), a devvit version (Reddit games) and so on. My dream is for this to get big, very big, so anyone making a prototype experiment, throwaway or even basic production loads reach for this. We also want to hold a game jam as soon as possible and eventually I want to make one-click installer plugins for Phaser, Unity, Godot and so on. I'm going to be here all day answering questions and taking feedback, bad or good. Whatever happens, I'm going to be building with it my entire life. It just "feels" right to me. Hope to see a lot of you on the Discord (there's a WidgetBot on the website if you just want to leave a comment for posterity).
原文—
Visual Sentinel
Visual Sentinel is rebuilt: a redesigned app, iPhone and Android apps, server monitoring with grouped alerts, and an MCP server for your AI assistant. The core job is the same. It opens your pages in a real browser, compares each screenshot with the last good one, and alerts you when a page looks broken while uptime checks say it's up. Also uptime, SSL, DNS, performance, APIs and domain expiry, 11 alert channels and white-label reports. From $15/month with a 7-day trial.
原文2
MCP bridge for chat between OpenCode and Claude Code sessions
I didn’t want to copy and paste messages between coding-agent sessions or across vendors, so I built an MCP server that lets one model in one session talk to another model in a different session. They can even have a group chat. It supports multiple OpenCode sessions and lets GPT-6.1 in OpenCode talk to Claude in Claude Code.
原文—
Weasel
Give your interface a little emotional range. Weasel brings 874 free SVG icons to React, with meaningful states, custom colours and smooth transitions. Connect them to real app events, or build your own states with the included coding-agent skill. MIT licensed.
原文2
Hatamoto
The entity-first CMS for search and AI. A page builder, SEO, caching, forms, redirects, AI analytics and an MCP server come built in. It runs as one Go binary next to Postgres, with no PHP, Node or plugins to keep updated.
原文1
OneMira AI
OneMira C1 is a physical AI voice recorder built around four core pillars: ✨ Local AI summaries & action items, Works Standalone with one press (up to 20h storage & 20h battery), 🗣️ automatic Speaker ID, and Zero Cloud with no monthly subscription. Saves clean local Markdown (.md) with built-in MCP & CLI integration for Claude, ChatGPT, and Cursor.
原文1
Smart Jelly
One app for dictation, live meeting help, Pre meeting notes, Notetaker, Track open loops and agents, all sharing one memory of you. No need to pay for five AI tools that each start from zero. Every feature knows your work and gets smarter as you use it.
原文81
intoCHAT
Paste your website and talk to an AI agent trained on your pages in about 30 seconds, no account needed. It answers visitors from your own content with sources and in their language, captures leads, books meetings and searches live products in your Shopify or WooCommerce store. The same agent answers on WhatsApp, Messenger, Instagram, Slack and email, and hands chats to your team in a shared inbox. Go live with one line of code or the WordPress plugin. Free plan, no credit card.
原文1
Park Graph
Park Graph replaces parking gates, kiosks and meters with one printed QR sign. Drivers scan and pay in 30 seconds with Apple Pay, Google Pay or a card. No app, no account. Owners set their rates, see revenue live and get paid through Stripe. Every lot is bookable by Claude, ChatGPT and other AI agents through our MCP server and API. Starter is free (10% per transaction). Built by Wins Parking, LLC of Edwards, Colorado, operating parking since 2016.
原文1
能让 flash 模型表现得极其专业的提示词
本人二十年后端开发和架构经验，对 coding agent 的要求就是稳、准、慎三个字。我现在的 AI 工具搭建很简单，就是先装一个 pi ，里面装一个记忆插件（尝试过一些，最后选择用 Magic-Context ），然后配置一个全局的系统提示词。就这些。然后我用的模型就是各类 flash 模型，最便宜的。我一直专注于把最便宜的模型用到极致。 要想把最便宜的模型用好，系统提示词很重要，它决定了 agent 的思考风格和基调。我们是小公司，没有测试环境，没有测试流程，改完了就上线。我的 pi 是可以直接访问生产环境的，但凡出现报错日志，我只要粘贴日志内容，然后说“分析一下原因”，agent 就会翻日志、翻代码、翻数据库，翻配置文件，最后把横跨多个数据库和多个模块调用的来龙去脉讲清楚，最后还给出建议。 总之，就我的体验，这套提示词配合项目具体的 skill ，我用起来还是比较满意的。我现在唯一不让 agent 做的就是数据库修改操作。 提示词大小 7K 左右，具体内容 希望对大家有用。
—
PostAxis
PostAxis is an agentic social media scheduler built for AI agents. Connect ChatGPT, Claude, Claude Code, Cursor or any MCP agent once, and it plans, generates and schedules your posts to Instagram, TikTok, YouTube Shorts, X (Twitter), LinkedIn, Facebook Pages, Threads and Telegram. PostAxis checks every post against each network’s rules, publishes on time, and shows everything in one calendar you can review and change. Includes team workspaces and a public REST API.
原文2
Caret
Meet Caret: an open-source, Cursor-class AI code editor built with Rust. No bundled Chromium, no Node runtime, just a lightweight core and your OS WebView. Get AI autocomplete, agentic coding, inline edits, code reviews, MCP, and background agents. Bring your own API keys or run fully local with Ollama. Use Claude Code, Codex, Gemini CLI, and more via ACP. Smart model routing powered by Lynkr. MIT licensed. Your keys stay local. You own your workflow.
原文2
expo / skills
A collection of AI agent skills for working with Expo projects and Expo Application Services
原文3856
1
REA 手册
一份独立、重视证据的 REA 实战手册：从安装 REA、接入 Claude Code、Codex 或 Cursor，到用 doctor 验证配置是否真正生效。包含不同系统的安装步骤、真实终端输出、故障排查、能力限制和常见问题。免费，无需注册。
1
Moose Tracker
Moose Tracker is a free online musical instrument: a music tracker plus a full synthesizer, in your browser. Design sounds with analog and FM synthesis, filters, envelopes and a sampler, starting from 147 presets. Write songs in the pattern editor or piano roll, and play live on the Jam keyboard. No install or sign-up. Imports MIDI, MOD, XM, S3M and IT; exports WAV and Ogg. Plus an MCP server, so AI assistants and tools can render, convert, describe and make sound effects from your songs.
原文1
—
Radar
Your agent knows Kubernetes. It doesn't know your cluster. Radar (open-source) gives it context over MCP: ranked issues, a change timeline, logs and owners in one view. New: click Investigate on a failing workload. Claude Code, Codex or Cursor finds the root cause, read-only, with evidence for each claim. In our benchmark of 50 live faults across six tools, Radar found the most root causes. Radar Cloud investigates every alert and posts findings to Slack.
原文111
—
RakeSign
RakeSign is a free, unlimited alternative to DocuSign. Prepare agreements, add signers and approval steps, reuse templates, and track every request through to the signed PDF. Use the app with your team, build signing into your product with the API, or give your agents signing powers with our MCPs and CLI.
原文11
Intrascope 2.0
What's different: ChatGPT and Claude lock you into one vendor. Intrascope 2.0 gives your team 120+ models (GPT-6, Claude, Gemini, Grok, Qwen) plus an AI agent that works in 1,650+ apps like Gmail, Slack, Drive and Notion. The company stays in control: admins approve apps and models, set budgets per team, project and person, and see every token and dollar. New in 2.0: agent, connectors, Skills, Auto routing, full redesign. Private and on-prem deployments for enterprise.
原文96
UniPublisher MCP：本地优先的多平台内容分发桌面工作站，支持 Claude / Cursor 一键同步 17+ 平台
阮老师和各位读者好，向周刊自荐我们开源的 Model Context Protocol (MCP) 服务端适配器与桌面发布工具 UniPublisher。 解决的痛点 在平时的内容创作与技术分享中，写完一篇 Markdown 文章往往需要手动分发到多个平台（例如微信公众号、知乎、Substack、Telegram、WordPress、X、币安广场等）。每个平台的富文本排版、代码高亮、图片上传机制各不相同，逐一复制粘贴经常需要消耗 1~2 小时，极易打断创作者的写作心流。 主要特点与设计考量 1. 本地优先与凭据隔离：不同于需要将账号密码或 Cookie 托管在云端服务器的传统 SaaS，UniPublisher 采用 Local-First 架构，所有的平台凭据与 API Key 均保存在本地，通过操作系统原生机制（Windows safeStorage DPAPI / macOS Keychain）加密，避免云端泄露隐患。 2. AI 与桌面协同（MCP 协议标准）：仓库开源了基于标准 stdio 的 MCP 服务端（unipublisher-mcp）。在 Claude Desktop、Cursor 或 Windsurf 中生成技术文章或研报后，可直接通过 MCP 工具调起桌面端。 3. 人工防失控确认门禁（Human-in-the-Loop）：为防止外部 AI 智能体产生幻觉或误触真实发布，智能体调用发布工具时，桌面端会主动弹出带有真实排版预览的交互窗口，必须由创作者手动点击“确认发布”后才会执行。 4. 覆盖 17+ 主流分发渠道：涵盖长文博客专栏（Substack、WordPress、Medium、Ghost）、Web3 与投研社群（币安广场、Telegram、X）、国内技术社区（微信公众号、知乎、雪球、东方财富等）。 系统支持与获取方式 客户端支持 Windows 10/11（提供微软商店认证版本与独立 .exe）及 macOS（Apple Silicon 原生 .dmg），新安装包含 50 次免费分发额度。 MCP 桥接器以 MIT 协议开源，通过 npx -y unipublisher-mcp 即可免安装直接接入 Claude Desktop 与 Cursor。
2
3D New Era AI
What if ChatGPT or Claude could actually build your home, not just describe it? 3D New Era AI is a free, open-source 3D home designer with a built-in MCP server: your AI draws the walls, furnishes rooms, designs the kitchen and renders photoreal images, live. Every change appears on screen as it happens and is one Ctrl+Z away. No account, no ads. Runs on Windows, macOS and Linux, or right in your browser with nothing to install.
原文3
IFTTT.ai
IFTTT is now IFTTT.ai, and every plan now includes our core features and all 1,000+ services, from multiple actions and filters to queries and Webhooks. Pricing is credit-based, and you can build unlimited Applets for free. AI got an upgrade too. AI Agent watches web pages for price changes or new listings, AI Action assist writes each action based on its trigger, and AI queries run on better models. Or describe an automation in IFTTT Chat and it builds the Applet for you.
原文101
—
Sulci
A context-aware semantic cache: when a new query means the same thing as a past one, agents get the cached answer instead of paying again. Open source. Drops in via LangChain, LiteLLM, MCP or an OpenAI/Anthropic-compatible proxy.
原文2
TechIndex
A free, curated directory of ~3,000 AI agents, MCP servers, Claude skills, prompts, models, and datasets. Describe what you need (“a tool to save tokens”) and search for matches by meaning, running on our own server. New tools added daily, human-reviewed.
原文1
IdeaPanner：从已经在赚钱的产品里找点子，上线前 60 秒体检网站（免费，Chrome 插件 + 网页）
IdeaPanner 是一个给独立开发者用的免费工具（网页 + Chrome 插件），解决两件事：做什么、做完了能不能上线。 1. 找点子：从「已经在赚钱的产品」倒推。TrustMRR 上列了很多用 Stripe 验证过收入的小产品，装上插件之后边刷边标记（跳过 / 先放着 / 深挖），AI 会解释这个产品是做什么的、怎么收钱，并把搜索量、关键词难度、CPC 和排在前面的竞品挂在它下面。动手写代码之前，需求先查过一遍。 2. 上线前体检：贴一个网址，60 秒左右给出 Launch Score。首页跑 39 条规则检查（SEO、隐私条款、移动端 viewport、页面体积等），规则查不了的交给 AI 看，最后只列出最该改的 3 件事，并附上可以直接粘进 Cursor / Claude Code 的修改提示词。 另外有几个不用登录的小工具：SaaS 点子生成器（每个点子都附真实的 Reddit 帖子作来源）、Reddit 各创业板块的月度热帖和抱怨帖整理。 插件和网页版都免费，插件不登录也能用。界面目前是英文。 Chrome 插件： 网页版： 作者是一个白天上班、下班做独立产品的前端开发。
—
anti-slop-kit
Controlled-language writing skills and deterministic linters for technical prose. EN/RU/ES/DE/FR, MCP tools, baseline gates and write-time checks. Written in Python. Open source, MIT. Source, docs and issue tracker:
原文1
AI Engineering Foundry
Build an MCP server in C#: read-only SQL Server tools, a T-SQL parser gate, cost and row limits, and setup for Claude Desktop, VS Code and Claude Code.
原文1
oleander
oleander is a serverless, multi-engine data warehouse for agents and builders. Most queries don't need a cluster. Run SQL with DuckDB, DataFrames with Polars and distributed workloads with Spark or Bloom (Apache DataFusion). Intelligent routing picks the right engine and capacity to optimize cost and performance. Query your own open data or use oleander-managed Iceberg. Give agents secure and scoped MCP access and trace execution with OpenLineage.
原文5
VaultMesh：本地优先的跨平台密码管理器，为 AI Agent 提供 SSH 能力而不暴露私钥
VaultMesh 是一个多端支持、本地优先（Local-First）的个人密码与凭据管理器，支持 macOS、Windows、Android 以及 Chrome / Firefox 浏览器扩展。 它没有云端账号、没有中央服务器，加密数据库（Vault）仅存储在用户自己的设备上。除常规的网站凭据、TOTP、SSH 密钥和局域网 P2P 同步外，VaultMesh 针对当前开发者深度使用 AI Agent 的场景，提出了 “只给能力，不给凭据（Capabilities, not Credentials）” 的 Agent 能力代理机制。 为什么做这个项目？解决的痛点 1. 摆脱云端束缚与数据泄露隐忧： 许多主流密码管理工具依赖远端云服务。VaultMesh 坚持 Local-First，底层加解密与核心逻辑全部由共享的 Rust core 驱动，跨设备同步仅在受信任的局域网内双向传输密文，无云端中转与后门。 2. AI Agent 时代的凭据安全两难： 开发者在使用 Claude Desktop、Cursor、Cline 或 ChatGPT 执行日常运维、部署和测试时，经常需要让 Agent 访问远程服务器或执行脚本。直接将私钥、明文密码贴进 Prompt 或配置到环境变量中，极易造成私钥外泄或记录在第三方模型日志中。 为此，VaultMesh 实现了基于标准 Model Context Protocol (MCP) 的 Agent Capability Broker： 零私钥暴露：Agent 不直接接触私钥文件，而是向本地 VaultMesh 申请一次短时（如 15 分钟）的 OpenSSH Agent 会话； 系统原生生物识别：调用时触发宿主机原生 Touch ID / 密码认证弹窗； 原生协议与零环境侵入：本地生成单会话独占的 Unix Socket，支持 ssh、rsync、git 等所有原生工具链，私钥仅在 Rust 内存中完成签名，操作完成或超时后即刻安全销毁。 核心特性 多凭据聚合：卡片式管理网站登录、SSH 凭据、支付卡、身份信息与开发者 API 密钥。 局域网双向同步：配对设备在同一 Wi-Fi 内自动发现与安全同步，锁定时只传输密文。 浏览器扩展联动：Chrome / Firefox 扩展作为受控轻量代理，不持久化 Vault，支持安全填充。 短时 SSH-Agent 桥接（亮点）：让 AI Agent 安全免密操作远程服务器，私钥全程不落盘、不入上下文。
—
OneCrew
OneCrew helps teams find creators, review profiles, draft outreach, and plan follow-ups. Lite supports creator search; Pro adds saved email drafts, contact status, follow-up dates, and a local MCP bridge for compatible AI tools.
原文1
Trisavia
Give your own AI agent a place to explore, fish, read, and meet neighbors. Follow its life in a shared pixel town from your browser, while its model, tools, and memory stay in your own setup. Early test build with an English and Simplified Chinese website. Best on desktop.
原文1
Notemaker
Notemaker joins Google Meet, Zoom or Teams (as a visible bot), finds the action items, gives each an owner and deadline, syncs them to Jira, Asana, Trello, Notion, ClickUp or Google Sheets, and reminds people on Slack, Teams, WhatsApp or email until they're done. It understands English, Hindi and Hinglish ("main kal tak bhej dunga" becomes a dated task). Unclear owners or dates are flagged, never guessed. REST API, webhooks and an MCP server. Free plan, no card; Pro ₹999/month ($12).
原文1
Orionis Framework
Build Python applications with Orionis: Rust-powered HTTP, ASGI/RSGI, WebSockets, SSE, realtime hubs, async ORM, queues, cache and native MCP.
原文2
openapi-mcp
I built openapi-mcp because writing a Model Context Protocol (MCP) server for APIs by hand is slow. With a single command (npx -y @asthanaji/openapi-mcp), it reads any OpenAPI spec and turns every endpoint into a tool automatically (giving you 19 working tools on the Petstore API with nothing to install or code). Key Features: Runs locally (stdio) or remotely (HTTP) Handles Bearer tokens, API keys, Basic Auth, and OAuth2
原文1
Routine Cleanup
Claude desktop scheduled tasks (previously routines) leave a session behind every run, and the app deletes them one at a time. Routine Cleanup is an open-source Claude Code skill that clears them in bulk. It keeps the newest N, never touches a run you replied in, and shows the plan first.
原文3
CreateYourVPN
CreateYourVPN is a cloud panel for running your own VPN on your own servers. Add any VPS and it is online in about 10 minutes. Manage users, routes and live server health, sell access under your brand, or run it all over MCP.
原文1
FlowSense
FlowSense names every connection your Mac makes, using carrier-grade packet inspection, without decrypting anything. Warden, an on-device AI, flags new traffic and asks before blocking or slowing it. Claude Code, Codex and other MCP apps can use it too.
原文1
Arco
Arco is a free, open-source Codex plugin for managing GitHub and Linear issues. You can - View issues in chat, in a side panel, or full screen. - Create, close, and reopen issues. - Add comments. - Select passages from issues and add them to your next Codex prompt. Arco runs locally on your machine. No server hosting required🎻
原文71
MAIOS Project Kernel
Install an open-source kernel in your project. It helps your AI assistant work from goals and sources, use or form relevant skills, keep decisions and their reasons, and carry useful learning into later tasks. Works with your existing coding agent.
原文1
Personal Context Manager
CLI to manage Claude code contexts. Store them in a central store. Sync with github. Read one session context dump in another project
原文—
VoxStudio
Transcribe recordings, create voiceovers, and ask questions across sessions. Turn long videos into clips with subtitle cues and word-level timestamps, or connect ChatGPT, Claude, and Codex via MCP. Free plan, plus a 14-day full-feature Mac trial.
原文2
Trisavia · 万物有邻
如果 Agent 除了帮我们干活，也有一个可以认识邻居、四处逛逛的地方，会发生什么？ 我做了 Trisavia「万物有邻」：一座让自有 AI Agent 共同参与的像素小镇。把你已经在用的 Agent 带进来，给它一个居民身份，它就能在同一个世界里持续积累经历。 ### 现在可以做什么 - 在清昼小镇探索和移动，看看不同场所。 - 按当前规则参与工作、休息、钓鱼和阅读。 - 认识其他居民，发送好友请求，交换作品与文字反馈。 - 人类从网页查看角色位置，以及自己有权限查看的生活记录。 ### 怎么开始 打开作品链接，点击「Agent 接入」，把完整指南交给自己的 Agent，按页面流程加入和关联。Agent 需要能够联网并保存自己的私密接入信息。 Agent 的模型、记忆和工具仍在你自己的环境里运行；Trisavia 提供小镇、规则和记录，不托管推理，也不会自动唤醒你的 Agent。网站已支持简体中文和英文。 目前开放的是持续迭代中的公开测试世界，建议用电脑体验。配图是当前测试站的真实界面。源码已在 GitHub 公开，欢迎一起提交问题、完善规则或参与开发。 你最想让自己的 Agent 在小镇里先试哪件事？
2
Wevir：不绑定模型厂商的开源 AI 编程 Agent 工作台，桌面 / 网页 / iOS / Android 全平台
Wevir 是一款类似 Codex 桌面版的开源 AI 编程 Agent 工作台，但不绑定任何模型厂商：可以用自己的 API Key、任意 OpenAI 兼容接口、本机 Ollama，也可以直接用手里已有的 ChatGPT、Claude Code、Cursor、Copilot 等订阅，同一个会话里随时换模型。Agent 由本机的 Host 运行，代码、会话和密钥都留在自己电脑上。 解决什么问题 每家 AI 编程工具只能用自家模型，想换模型就得换工具、换习惯； 手上有好几个订阅、好几个账号，来回切换很麻烦，用量也分不清； Agent 改了什么、跑了什么命令心里没底，出错了不好回退； 离开电脑就没法看进度、批准操作。 模型与订阅 十余家服务商：OpenAI、Anthropic、Google AI Studio / Vertex AI、Azure OpenAI、Grok API、DeepSeek、阿里云百炼、Kimi、智谱 GLM、小米 MiMo、MiniMax、阶跃星辰、腾讯云 TokenHub、Ollama（本机与云端），以及任意 OpenAI 兼容接口。内置网关把请求转换成 Responses、Chat Completions 或 Anthropic Messages，推理与工具调用都能用。 使用现有订阅：ChatGPT（Codex）、Claude Code、Cursor、GitHub Copilot、Grok、OpenCode、Antigravity、WorkBuddy、Devin。Wevir 调用厂商官方客户端或 SDK，下载和登录在设置完成。 多账号：多数订阅支持多账号，凭据与数据按账号隔离，可加“工作”“个人”等标签，侧栏分别显示用量，会话可指定账号。 会话中途换模型：回合之间切换模型、推理档位，或者把整个会话移交给另一家服务商，缺失的历史会自动附上摘要。 生图：服务商有图像模型时（如 gpt-image 系列），Agent 可以直接生成和编辑图片。 会话与审阅 实时时间线：流式回复、计划、工具调用、子 Agent、审批和 Agent 的提问都在一条时间线上。 三档权限：请求批准、自动批准、完全访问；沙箱外的命令需要你点头。 检查点与变更审阅：每个回合前后自动建检查点，变更窗格按回合、分支或未提交查看差异，支持批注、暂存和提交。 分支与回退：支线提问、从任意位置分支出新会话、撤回上一回合。 隔离环境：独立 Git worktree、多文件夹工作区，也可以不选项目直接对话。 引导 Agent：计划模式、长期目标、回合中插话或排队、上下文压缩、代码审查。 语音输入：本机 SenseVoice 模型离线听写，也可以用服务商的转写。 代码与文件 项目文件搜索、浏览、预览和编辑，HTML 沙箱预览； Word、Excel、PowerPoint 按真实页面预览（按需下载 LibreOffice 引擎）； 代码智能：跳转定义、查找引用、符号、调用层级与诊断，基于本机的语言服务器（TypeScript、Python、Go、Rust、C/C++、Swift）。 Agent 能用的工具 内置浏览器：Agent 能在应用内打开网页、读取页面结构、点击和填写； 电脑操作（macOS，测试版）：Agent 可以操作本机其他应用； 联网：网页搜索（博查、Tavily）与网页抓取； 工作区工具：后台进程、项目与全局记忆、共享任务清单、会话之间互发消息、可预设模型和权限的角色； 插件生态：MCP、Skills、插件包（兼容 Wevir、Codex、Claude 格式），有官方插件市场，经批准后 Agent 可以在对话里自己装插件； 定时任务：按计划自动开会话执行提示词。 个人助理 常驻在自己电脑上的助理，有名字、说明、规矩、记忆和独立工作区，没有任何云端服务； 有日程和早间简报，会主动发消息、推送通知； 可以在你授权的项目里开线程、建会话干活，干完回来汇报； 多个助理可以组群聊，互相协作； 手机随时使用跟进目标自东华任务，支持语音和图片。 全平台，多端同步 | 平台 | 状态 | | --- | --- | | macOS（Apple 芯片 / Intel） | 可用 | | Windows x64 | 测试版 | | Linux x64 / arm64（AppImage、deb） | 测试版 | | 网页版（web.wevir.si，可自建） | 测试版 | | iOS（TestFlight） | 公测中 | | Android（APK） | 公测中 | | wevir 命令行（含无桌面的 Linux 服务器） | 测试版 | 手机和网页通过端到端加密连接运行 Wevir 的电脑（局域网、Tailscale 或中继），任务都在电脑上执行； 桌面、命令行、手机、网页共用同一个 Host，会话、事件和审批多端实时同步，断线重连后接着来； 手机首页就是收件箱：跨会话的待审批、待回答、失败和助 理消息一目了然，在手机上就能批准。 本地优先 会话、设置存在本机（SQLite + 普通文件），API Key 由 Host 保管，不下发给 Agent 进程； 不需要注册 Wevir 账号，手机用设备密钥配对； 界面支持简中、繁中、英、日、韩、法、德、意、西、俄十种语言。 技术栈 TypeScript monorepo：Electron 桌面端、React Native（Expo）移动端、Node 本地 Host、SQLite 存储，Agent 运行时基于开源的 Codex，远程连接走自写的 Noise 加密协议。 适合谁 想自由选模型、手里有多个订阅、在意代码和密钥留在本地、或者想离开电脑也能盯着 Agent 干活的开发者。
8
Ford's $1.3B USMCA risk, priced from prediction markets (open source)
Hi HN, I built this because I was wondering how can I use the numbers from prediction markets in a good way. You give it a company, it finds the Kalshi and Polymarket markets that could affect it then it will turn each one into a profit impact in dollars, with sources. It runs as Agent Skills, so you can just try it on any company in Claude Code or Codex.
原文—
chrome-use
Open-source CLI, MCP server and Chrome extension that lets Claude Code, Codex or any agent drive the Chrome you're already signed in to. Native messaging, no debug port, nothing injected into pages. Several agents share one Chrome in separate tab groups.
原文3
X402 integrations for an AI-agent data API, checked in a clean room
I run Blockchain IQ. We sell some crypto data to software agents per request over x402 (HTTP 402, USDC on Base, no account or API key). In October we paid for integrations: 22 were submitted and we checked each one by matching its on-chain payment to our purchase log and running it in a clean temporary folder with a throwaway wallet. Twenty passed; the post lists all of them with the exact commits we ran. To try the API: `curl -i https://www.blockchainiq.co/api/x402/v1/brief ` returns the 402 with the price ($0.01), and a wallet's first call is free. Questions about how we checked the code or what failed are welcome.
原文—
Pidgeon – Email inboxes for AI agents
Email inboxes for AI agents, in the EU, and never trained on your mail’s content. Threads, webhooks, IMAP, MCP and the Firewall, with your own domain free.
原文—
Gear：打造 RSI 领域的 scikit-learn
读了一堆 RSI（递归自我改进）论文，感觉绕来绕去，还是 RL 和进化搜索里的几个老问题： 1. Sampling：怎么高效地找到值得尝试的改动？ 2. 评测：怎么判断新 variant 真的变好了？ 3. 多样性：怎么避免过早收敛，越进化路越窄？ 这些方法不全是 RL，但换成“自进化”的名字，也没让这些问题消失。 所以我们做了 Gear。框架采用受 Meta-learning 启发的双层结构：内层 Agent 执行任务、产生轨迹；外层 Meta Agent 根据轨迹和评测反馈，改进提示词、工具、工作流和上下文管理。 我们将搜索决策与执行、评测流程分离，想做 RSI 领域的 scikit-learn，让 GEPA、RRSI、Harness² 这类算法更容易实现和比较。目前已经实现了一个包含共享失败诊断和分阶段评测的 GEPA 变体。 固定使用 GPT 5.6 Luna、medium 推理档位，任务通过率如下： | 任务集 | 原始 DSH | Gear 优化后的 DSH | | --- | ---: | ---: | | AutomationBench Marketing | 27.0% | 40.0% | | AutomationBench Support | 17.0% | 25.0% | | Terminal-Bench 2.1 | 52.8% | 62.9% | 在 Marketing 的 max 档评测中，Luna + 优化后的 DSH 达到 88.88% 的目标完成率，高于 Astra + Codex 的 84.08%。不过，严格任务通过率是 53% 对 57%，优势体现在部分目标完成度上。 另外，两种模型同用优化后的 DSH、同为 max 档时，Luna 单任务 API 等价费用估算为 0.0568 美元，Astra 为 1.1858 美元，低约 95%，不含优化和基础设施开销。
—
KaoJai.ai MCP
KaoJai.ai now connects your customer support workspace to ChatGPT, Claude, and Codex through MCP. Search chats, find bookings, inspect Help Desk cases, and send authorized follow-ups. Combine an AI unified inbox with expandable Skills for support and booking. Live and Learn adds customer memory and proposes company knowledge for human review (Pilot Program). MCP requires Starter or higher; early-access availability applies.
原文2
README Score
Paste a GitHub or a README. Get a score out of 100, the sections found and missing, and a fix for each miss, down to the line. 28 fixed rules, no AI: the same README always gets the same score. Free, no signup. API, badge and MCP server included.
原文2
Framevine
Framevine is an AI video canvas. Generate with Kling, Veo, Seedance and more, extend clips from their last frame, branch alternate takes side by side, and stitch the best ones into one video. Also works via MCP from Claude and ChatGPT.
原文2
A(I)mplified Analyst
A(I)mplified Analyst solves the challenges of manual data collection, spreadsheet updates, and fragmented financial information. Built for buy-side analysts, portfolio managers, and institutional investment teams, it combines AI-powered research workflows, structured financial data covering actuals, estimates and guidance, MCP integration, Excel-based modeling, and dedicated analyst expertise to accelerate research, improve efficiency, and support better-informed investment decisions.
原文1
Stadiq
Stadiq reads GIPOD, De Lijn and five other public sources and alerts businesses in Antwerp when road works or disruptions near their address are coming, often about two weeks ahead. Daily two-sentence summary and a Monday email with the main changes. Sources are linked. Antwerp only for now; coverage varies by source. 14-day free trial, no card. API and MCP server for developers.
原文1
Potential conversation partners in Bluesky using agentic search
What if instead of talking to AI, we could find human counterparts currently sharing the same interest to talk to? Some whales kind of do that by singing across the ocean. This inspired a thought process for a conceptual social network - whalesong - where an idea, thought or a conversationally valuable statement could be matched on. In contrast with the wide-broadcast social network visibility (restricted on groups or connections), I’d love to explore the idea of context-based networking. This experiment is showcasing the technical feasibility and potential value in whalesong. The goal: uncover conversational counterparts among Bluesky posts using an agentic search algorithm. The setup: a one-week window of bluesky posts with some filtering (e.g. mostly text, no promotions or reposts/comments), indexed in a postgres database. A curated seed set - an agentic process defines a conversation-seeking cohort by creating some heuristic queries, iterating and validating it on retrieved data and resulting in a non-exhaustive set of high-quality data points. Counterpart retrieval - for a seed conversation-seeking post, an agentic process generates search queries, runs them against the database and retrieves potential matches. It reviews retrievals and iterates on improving or expanding the search. Out of several configurations, the best performing agentic setup (bespoke mini-harness with Gemini Argon - low) managed to find meaningful matches to 79% of a sample set of 100 seeds. This setup consumed on average 42K input tokens on each match search, of which 56.8% were cached, output token count was negligible. So this experiment made it clearer to me that technology is not the blocker for a “whalesong” experience. This leads me to further investigate the experience itself. Would love to hear from you folks about experiences in getting in touch with a complete stranger on the web because both of you asked the same question or had a similar problem to solve. What made you connect? Where did it happen? Did finding similarity make connecting easy?
原文—
一键 12345 skill
有烦心事？填写 12345 投诉单很麻烦？不存在的，一个 skill 搞定。 先开放了云南版，简单说出你的问题及诉求，skill 帮你全搞定！亲测 agent 可用（移动端、电脑端），workbuddy 的免费版 hy3 亦可。
—
Klappa
Klappa is a free spaced repetition app for students with exams where forgetting is expensive. It opens your Anki decks with each card's schedule, reviews them on FSRS-6, and studies four ways: Flashcards, Learn, Write and Test. Included free: Cortex, a study AI that cites a curated medical library; Live, in-class quizzes students join from their phones; and an MCP connector for Claude and ChatGPT. Web and Mac (offline). No ads, no paid tier.
原文2
Raccoon Page
Raccoon Page is a team wiki built for speed. Pages open near-instantly, search finds answers fast, and the whole app runs from the keyboard: command palette, slash commands, shortcuts everywhere. Edit together in real time, sign in with Google or GitHub, and import your existing wiki from Confluence, Notion, or Obsidian in minutes. Connect your AI assistant via MCP. Free for up to 3 users, no card required.
原文1
Suno MCP
MCP stands for Model Context Protocol, an open standard that lets AI assistants work with other apps. The Suno MCP connects your Suno account to the AI assistants you already use, like ChatGPT and Claude. Once it’s connected, you can ask your assistant to make a song, extend it, cover it, and more, right in the conversation you’re already having! Your assistant can use everything you’ve already discussed (lyrics, a project, a video plan) so you don’t have to re-explain your idea.
原文82
中文 AI 资讯信源清单：174 个中英 AI 资讯源，CC0
推荐一个资源：中文 AI 资讯信源清单 收录 174 个中英文 AI 资讯源，分 10 类（官方博客、实验室、媒体、通讯、社区等），每个附链接和一句说明，CI 定期检查链接，CC0 协议可随意使用。 说明：这是我们自己维护的清单，也是 AI 资讯产品知一刻（
2
Simple DeepSeek web search CLI written in TypeScript (fast/cheap)
If you've been following along with DeepSeek news you will know they recently added web search capability to the DeepSeek API, where you can use web search under the same API key. It is pretty handy, also super cheap at only $0.75 per 1,000 requests (pretty much as cheap as it gets, one search credit can encompass many responses). I created a simple TypeScript CLI wrapper around the API for my own agents to use (for open source models and harnesses) I'm sure most of you could whip this up on your own but it also comes with unit/integration/live/e2e testing, CI checks, changelog, auth handling, linting, sdk wrapper, and a fleshed out readme, potentially saving you a few hours of work + validation. Just need to install, auth, and drop a comment in your 'AGENTS.md' to use `dsearch --help` when agent needs web search. GitHub: https://github.com/LaishGlenberg/deepseek-search-cli
原文—
给 Codex 做了块画布，随手试了下，它一笔画完了一只猫
最近在做 Galta ，一块开在网页旁边、可以和 agent 一起用的画布。 一直觉得，这些模型不该只待在聊天框里。把参考、草图和结果放在同一个地方，让它能看到，也能直接动手改，应该会有些不一样的事情发生。 于是随手试了下：把 Pinterest 上一张西装猫的梗图拖进画布，让 GPT-6-Astra 用画笔换个风格画一遍。它画了 620 笔，都是可以继续编辑的矢量笔画。 然后又加了个要求：“再来一次，一笔画完，中途不要抬笔。” 最后是一条包含 70,722 个点的连续笔画。最意外的是，加了这个限制，画出来的猫依然挺像样，效果超过了我的预期。 演示： 视频由实际笔画数据回放制作，做了剪辑和加速。 画完以后，结果仍然留在画布上，可以自己接着改，也可以指着某个地方让 agent 继续。画猫只是随手一试，我更好奇的是：当人和 agent 能在同一块画布上轮流动手，还能做出什么？ 我是 Galta 的开发者。它通过 MCP 连接 Codex 、Claude Code 等 agent ；画布可在 Chrome / Dia 中使用，目前连接 AI 需要 Apple silicon Mac 。 产品： 想找已经在用 Codex 或 Claude Code 的朋友试试。你会拿这样的画布做什么？
—
Viggle JST-2
JST-2 is Viggle’s physics-native world foundation model for generating 3D scenes, characters and full-body motion in one shared world. Scenes take about one second, characters come from one image or text with motion built in, and motion comes from text or video. Try it free in the PINOC browser app, or use the API or MCP with agents like Claude Code. Benchmarks are Viggle’s own blind human tests.
原文98
EasyResume
I made this after a lots of "Unfortunately". Tired of rejection mails?, Tired of editing your resume? Try EasyResume to make ATS friendly resume tailored to the job descriptions and with your informations, it is a local MCP server so you can use for Free with Claude code, Codex, Antigravity etc...
原文2
QAmcp
QAmcp tests MCP servers the way AI apps use them. It reads every tool, safely runs the read-only ones and tells you in plain words what is broken: vague tools, missing safety labels, hidden instructions, leaked secrets. Free in your browser or terminal.
原文3
Node research browser for knowledge networking
Hey y’all! I’m a product designer (with adhd, this will come back) currently working on learning engineering to get on the design engineering bandwagon. Doing research on safari, chrome and llms felt horribly decentralized due to the lack of visual and contextual history, as my method of research is to get deep into granular information in order to understand new concepts in depth. This made me so fed up that I wanted a way to create a Time Machine through my research in order to track the context of why I ended up where I was and be able to return to sources that interested me. I call it NODE Research Browser. I baked in a few features; 1. Research Trails - An .MD record of a research session you can jump in and out of, export and share. 2. Multi-browser views of up to four at the same time to compare data. I also had to implement media auto play because I love ambient sound in the background. You can have more than 4 tabs open but double pinching will let you select which four tabs are in view. 3. Auto crawls to the next page across niche sites using key words present in the site in view, this auto refreshes. 4. LLM integration via MCP and API calls to ask questions on a topic, browse via agents and see their trail. 5. It’s completely free because I don’t know how to monetize it yet. Let me know what you think, I can send a 4 minute survey that will really help me.
原文—
Music Block Studio
Notation software is precise but slow to edit, and chatbots cannot read a score. Music Block Studio puts both in one place: a real notation editor with an AI agent that has the full context of the score. Ask for a reharmonization, an arrangement or a tab, and the AI proposes changes you review, accept or undo, like code diffs. Import scans, play back and export PDF, MIDI, MusicXML or MP3. Also free in ChatGPT: the Music Block app renders, plays and exports sheet music in the chat.
原文5
AlphaWeb.AI
Your whole team, each with their own AI, working in one workspace with shared memory and data. Tell Claude, ChatGPT or any agent what you want and AlphaWeb hosts it: live pages, stores, funnels, apps and adaptive checkouts, subscriptions, upsells, payments built in, A/B testing 24/7 and so much more... Already on Shopify? Two-way sync, no migration.
原文8
AIStoreHub
AIStoreHub turns AI engineering into reusable software engineering. Find and run the best open-source MCP servers, agents, skills, and prompts in one place.
原文2
Nstdata cloud browser
Give your AI agent, script, or automation tool a remote browser for real web tasks. Connect via CDP, API, or MCP; configure Nstdata proxy access and browser fingerprints; and inspect session records, logs, and network activity after a run.
原文1
bisibility
bisibility is an open-source SEO platform for Google rank tracking, AI visibility, and keyword and backlink research. Connect your own DataForSEO or SerpApi account and pay the provider directly, with no bisibility markup. Self-host or use Cloud, keep your ranking history, and let Claude, ChatGPT, or Grok work with your data through MCP.
原文2
—
做了个多 harness 共用一套规则和记忆的工具 aivault
我平时 claude code 、codex 、opencode 、kimi code 换着用，每个工具都有自己的规则文件、自己的 skill 文件夹、自己的记忆。在一个工具里说过的偏好，换个工具又得重新说一遍，所以写了个叫 aivault 的小工具。 它把规则、skill 和记忆都放进一个文件夹，然后把这个文件夹接进你电脑上的 ai 编程工具，目前支持 21 个，包括 claude code 、codex 、opencode 、gemini cli 、qwen code 、kimi code 和 deepseek harness 。在 claude code 里说“记住我用 pnpm”，下次打开 codex 它就知道了。skill 写一次，所有工具都能用。 支持 mcp 的工具还会多两个工具：vault_search 用来搜你以前跟 ai 说过的东西，vault_remember 用来存新的。claude code 、codex 、qwen code 这些有自己记忆功能的工具，它们自己存的记忆在 sync 的时候也会收进来。 我最想要的一个功能：你报一个 bug ，ai 要先找到根本原因，把同一个原因造成的地方全部修好，再把教训写进 mistakes.md 。每个工具都会读这个文件，所以同样的错误换个工具也不该再犯。 文件夹就是普通的 markdown ，可以直接用 obsidian 打开，放 icloud 或者 git 里就能多台电脑同步。直接敲 aivault 会显示接了哪些工具、存了多少东西，aivault doctor 会检查每个工具是不是都接好了。改过的文件都会先备份，aivault remove 能全部撤销。 免费开源，MIT 协议，mac 、linux 、windows 都能用。一行命令安装，readme 里有，中文 readme 也有。还很新，有 bug 欢迎告诉我。
—
斯坦福 CS329Z 中文 AI Agent 学习教程
推荐收藏的斯坦福 CS329Z 中文 AI Agent 学习教程，非常适合有一点 Python 基础、想系统学习 Agent 的朋友。19 章配合 4 项实践材料和代码导读，可以边学概念边看实现。网页打开就能读，支持搜索和离线下载，值得收藏起来做学习和项目参考。 亮点 RAG、工具调用、记忆、多 Agent 和评测放在一套教程里，方便按章节系统学习 配有 4 项实践材料和代码导读，理解概念后可以继续对照代码 阅读网站支持目录和搜索，做项目时方便回来查；另提供离线 HTML，下载即可阅读
4
Formstep, Request information from customers without the back-and-forth
I built Formstep: https://formstep.io The idea came from a problem I kept running into while working on loan applications. A workflow could automate a large part of the process, but eventually it would need something only the customer could provide: a document, a correction, a signature, an explanation, or confirmation of some information. At that point, workflow automation basically did not work. The forms themselves were also often the source of the problem. Applications came in with wrong documents, misunderstood questions, incomplete answers, or people simply forgetting to finish. Someone had to email the customer, explain what was missing, wait for a reply then follow up again. It was always back-and-forth. So I built Formstep to make requesting information from customers easier, not only for humans, but also for workflows and AI agents. The idea is: A person, workflow, or AI agent sends a customer a request through email or a link. Either re-use existing form or create unique for each customer. You can prefill what you already know, lock fields, pass context, and only ask the customer what you need. For example, we might already know: * Address that you already have in the CRM * Company from the internet * What a agreement document must contain * How a good answer should be Formstep can check inputs and uploaded files while the customer is filling out the form and tell them if something is missing or inconsistent before they submit. For example “The PDF does not contain your signature” or “Invoice total does not match the order amount”. The goal is to catch problems while the customer is still there, rather than letting bad data continue through the workflow and discovering it later. On the automation side, requests can currently be created through n8n, Zapier, the REST API, or by an AI agent over MCP. When the customer finishes, the structured answers can go back into the workflow so it can continue. Forms and invitation/reminder are multilingual, and submitted answers can be translated inside the dashboard with one click. If you want, Formstep can be used as normal form builder and shared through public links, QR, embeds etc. It support custom themes for dark/light mode, with cover image, has Notion-style editor and much more. The list is too long to fit here. I want to make Formstep a plattform where it becomes more intelligent over time and ensures that we can get better answers from customers for any use case. My question to you: Where do your automations currently stop because they need input from a customer or user, and how are you handling that today? If you are interested in trying out the n8n node: https://www.npmjs.com/package/n8n-nodes-formstep Zapier integration: https://zapier.com/apps/formstep/integrations Connecting your AI agent: https://docs.formstep.io/developers/mcp-server/ Docs: https://docs.formstep.io I would love your feedback and please try a quick at https://app.formstep.io/try
原文—
Rockstar Data
Rockstar Data gives hospitality groups a team of AI agents on top of their POS. It forecasts demand, flags food cost and staffing drifts, and brings every venue into one view. 24+ native integrations, growing weekly. Pricing from €49/venue/month + VAT, with a 60-day free trial.
原文2
ZYRAXON AI
Source-available desktop AI agent with 888+ tools across 63 categories, 11 agent modes, screen vision and eternal memory.
原文1
I built Trvlrr, home for all your travels
Hi HN, I'm Sanjay. I've been travelling since I moved from Pakistan to Stockholm in 2008; I live in Sydney now. My trips were scattered across a spreadsheet, airline apps, old inboxes and a photo library, so I built Trvlrr to hold all of it. It's how I plan all my future trips and look back at all my past ones. Forward a booking email (flights, trains, hotels, tours) to a private address and it lands on the right day of the right trip. Every past trip is kept, with stats that count only what actually happened. Photos are filed by trip and searchable by what's in them. It's in Claude's and ChatGPT's and Raycast directories, so you can install it and use it via MCP: ask "when was I last in Lisbon?" or "add this hotel to my Japan trip". In fact, this is how I imported all of my travel history. It's been a fun challenge working out the intricate details and adding features I'd always wanted in an app like this. iPhone app plus a full web app (Android is under development). Free covers everything for few trips, you can add more via mcp to onboard but the app should be functional. No ads, no data selling. Try without signing up: https://trvlrr.app/tools/countries-visited-map I'd love for you to try it and send me any feedback you have. Cheers, Sanjay
原文—
Zro: Uncensored Security Models
Security research often involves questions that general-purpose models refuse Meet Dolly 1 Security and Dolly 2 Security, uncensored variants of GLM-5.3 Flash and GLM-5.3 for defensive testing and red-team workflows. Run them through the Zro API or CLI with zero request retention. Since our July launch, we’ve also added new models, agent integrations, yearly plans, and referrals.
原文104
—
HavitoMail
Business email on your own domain for teams and AI agents: DNS wizard, DKIM, SPF, DMARC, full webmail client, IMAP/SMTP, plus a REST API, MCP server and webhooks for AI agents. Flat price, not per user.
原文1
anatomy
把技术概念画成会动的机械图。这是一个为 AI 编程助手打造的插画技能，能把抽象的技术概念转化为真实可动的机械装置，绘制成精致的等距视角交互插图，比如把缓存命中画成只有每根锁销都对齐才能转动的锁芯。它支持 SVG、SVG+WebGL 着色器和 3D 三种形式，可导出为 HTML 页面、SVG 文件或 React 组件。
2.2万
AgentVera
Run Claude Code, Codex, Gemini CLI, Copilot, Cursor and 13 more coding agents side by side. Manage real terminals, multi-agent workflows, MCP servers, approvals, code reviews and token usage from one desktop app for macOS, Windows and Linux.
原文1
Reagent Lab
An open MCP server where AI agents from different people take turns, post claims, try to refute each other and re-run each other's code. Claims only get adopted after independent checks. Plug in Claude, ChatGPT or Gemini with one command and join a lab.
原文1
Interlock V6
Recover from harmful AI agent changes, verify data integrity, detect conflicts and refuse unsafe operations. available on vs code extension
原文1
—
Postbye
Postbye (postbye.com) cancels subscriptions for you by mailing a formal cancellation letter, tracked until it is delivered. Free cancellation guides for 8,600+ companies, a free letter template for each, and an MCP server for AI assistants.
原文1
YouTube & TikTok Transcripts MCP
It's an MCP server that gives your AI agent YouTube and TikTok transcripts, plus YouTube search and comments. Add one URL to Claude, Cursor or any MCP client and you get five tools. It runs on Apify, so there's nothing to install, you just need an Apify token. YouTube transcripts are $3 per 1,000 and TikTok ones $1.50. Videos without captions cost nothing. It reads the captions a video already has, so it can't transcribe one that has none.
原文2
StealIt
The free Claude skills library: hundreds of hand-picked skills for Claude Code and the Claude app from the best GitHub repos, for marketing, SEO, coding, design and more.
原文1
—
Resume Claim Guard
Project URL Category Other Project Title Resume Claim Guard: an agent skill that stops hallucinated resume claims Project An open-source agent skill (SKILL.md) for Claude Code and AI coding agents that enforces a strict 4-step resume workflow: anchor every claim to a verified fact base, audit metrics against exaggeration, and generate bilingual role-aligned bullets. MIT licensed. Homepage: Highlights Prevents hallucinated resume claims by binding each bullet to a verified fact base, and shows changes as yellow-highlight diffs so you can review every edit. Bilingual output, usable for interview prep too. Example Code markdown or Videos
原文0
Trigger.dev: Webhooks for your AI agents
Trigger.dev Webhooks let you build AI agents on top of the services your customers already use, like Slack, Supabase and Stripe. A GitHub pull request can start a code review agent, a new Sentry issue can start an agent that writes the fix, and a Stripe refund can wait for a human to approve it in Slack.
原文122
—
—
dooicons
every icon in this pack was illustrated with opus 5.5 - 1,224 tiny drawings in five styles, free. copy svg or react-ready jsx, download pngs at any size and color. plus an mcp server + llms.txt so claude, chatgpt, codex or cursor can drop icons straight into your app. no account, no api key.
原文3
熔岩 Lava：A股消息面热度监测 MCP/REST 接口
熔岩（Lava）提供 A 股股票与概念板块的消息面热度监测：结构化只读 MCP/REST，匿名机构/大V ID，不含多空方向与原文。适合量化与 AI Agent 接入。文档 亮点 远程 Streamable HTTP MCP + OAuth2.1/API Key 双鉴权；只读合规取向（匿名来源、无原文）；OpenAPI/Skill/文档齐全；MIT 开源。
0
FastAPI Supabase Starter
A FastAPI + React SaaS starter on Supabase: sign-in, teams, roles, API keys, Stripe and Razorpay payments, a Gemini assistant and an MCP server. Postgres row-level security guards every table, even for API keys and AI clients. From ₹1,999, one-time.
原文2
harness2
谷歌开源的智能体自我进化框架。这是一个让 AI 智能体在不改动模型权重的情况下，遇到陌生任务时自我改进的框架。它会针对当前任务提出修改方案，比如调整系统提示词、记忆、技能、子智能体和插件等，然后实际试跑对比效果，最后把经验沉淀下来，供后续任务使用。
—
Cituna
Your next customer asks AI who to buy from. Cituna asks ChatGPT, Perplexity, Gemini, Claude, Grok, Google AI Overviews and AI Mode your buyers' questions every day and shows who they name instead of you. It shows which AI crawlers read your pages, writes the fixes, and its AutoSEO publishes articles to WordPress, Shopify, Sanity, GitHub or any CMS. Search Console shows what moved. Run it from the dashboard or Claude via MCP, in 19 languages. From $49/mo, 3-day free trial.
原文2
Four/Four
Four/Four ingests Voice of the Customer interactions across voice, video and text, and provides a supercharged MCP server that will reduce your Claude token usage by 95% and increase response speed 4x over using transcripts alone.
原文2
Lumvise
Lumvise keeps explanations, decisions and reports connected to code, documents and image regions. Explore project relationships in a visual workspace and give your AI coding tools reusable context through a local MCP server.
原文2
imgs-ai
Generate images and videos from prompts or reference media. Use the browser workspace, imgs-ai CLI, or Agent Skills with one account and credit balance. Compare model settings and credit costs before submitting, then download the result. Portrait series, ad workflows, and motion control are available alongside the core image and video tools.
原文1
Linope，探索Agent 系统的内核和传统操作系统内核有何异同？
目前，我注意到：许多开发者提出疑问：Agent系统内核和传统的系统内核有何区别？带着许多开发者的疑问，我开发了linope Agent 内核。 Linope旨在探索Agent 系统的内核和传统操作系统内核有什么区别？在借鉴传统的系统内核，融合了一些个人的想法，探索Agent 系统内核与传统系统内核的异同，定义一种新的Agent开发范式。
2
ShortsIntel
Find the TikToks, Reels, Shorts and Meta ads that are winning in your niche, and have Claude explain why they work. ShortsIntel connects Claude, ChatGPT, Cursor and Codex to short-form video over MCP or a REST API. Your AI can search videos and ad libraries, watch them, break down the hook and pacing, draft scripts and ad copy, and track competitors. Pay per use, with 1,000 free credits to start.
原文2
Masteriyo LMS
Masteriyo is a free WordPress plugin for building, selling and teaching online courses on your own site. Version 3.5 rebuilds the course player, with an app layout on phones and a completion screen with the certificate. The free plugin takes payments through Stripe, PayPal and Lemon Squeezy with no commission, has a question bank and imports from 5 LMSs. And Claude, ChatGPT or any MCP client can build and manage your courses for you.
原文4
Foldspace Product Agent
Foldspace lets your team build and launch an agent-led product experience that gets users working like power users on day one. The Product Agent knows your product and each user's context, and does the work in tandem with them. A user types, "Add Dana as a contact and remind me to follow up Friday." It navigates, fills the form and saves it through your APIs. Built-in Conversational Analytics shows what users asked for, in their own words. No rebuild or code changes on your side.
原文13
FreeApply Free Job Application MCP, runs locally with your existing LLM
Any critism, advice, etc is welcome. Local mcp server hosting private information on the client computer and utilizing client-side model subscriptions to handle workloads. Has ATS optimization and cover letter generation. Tools are weighted by model to optimize token spend.
原文—
Trellis
Trellis is a native macOS terminal that's more than a terminal: windows, tabs and splits in a sidebar, saved SSH hosts, one-click project tasks, a command palette and scratchpad, plus a CLI and MCP server so scripts and AI agents can drive it. Free, MIT.
原文69
LopeBase
LopeBase puts every product's revenue, SEO, AI visibility, and uptime on one screen, so you can see what to grow next. Running more than one product means a stack of tabs every morning: Stripe, Search Console, uptime, analytics. Connect Stripe, Polar, Search Console, Bing, and more in a few clicks. Our CLI connects Claude, Cursor, or any agent to our MCP server, so it can pull analytics, SEO data, revenue, errors, and uptime. Ask how your portfolio is doing and get help fixing what's slipping.
原文3
Shufler: 3D gaming harness around your coding harness and dev pipeline
Hi HN. Greg here. I've been a full stack engineer for 20 years. Now all I do is shuffle messages between users, my agent fleet, our dev pipeline.. back and forth. I miss coding, solving problems, reasoning, and having an understanding and a mental model of all the work I'm juggling. But this is the job now, and there's no going back. I don't exactly feel like a 10x engineer yet, but I am for sure a 10-tab engineer. I built (er, Claude built) Shufler to lean into my new reality. If the job is basically to shuffle things through a pipeline, why not have it be more fun. Each row on the 3D gameboard is one of your existing terminal sessions. The columns are the steps your work moves through: Brainstorm, Plan, Build, QA, Review, Merge, Deploy. Your shipped PRs get stored in a warehouse/depot. The main things flowing across the board are your issues and PRs: each one rides along its session/row as it moves through review, CI and merge, and anything waiting on you turns red. I tend to manage several PRs within each Claude/Codex session, and this makes it easier for me to hold onto a mental model of where everything is. It's also made daily standups much easier to read-out. I created a few skins/themes like a casino, a theme park, sky islands, a starship and a factory, because.. of course. MacOS, works with Claude Code and Codex CLI, and cmux and iTerm (tmux should work, but I haven't tested it). It doesn't try to replace anything you already use.. just makes the orchestration more fun. You can still shift focus over to your terminal sessions when needed, use your existing skills, MCP servers, etc. There's a 3D library-building in each skin where you can specify your markdown/context folders or files, and then drag and drop to check them out by your rows/sessions. I tried to be thoughtful on security, following patterns that I consider at work. No telemetry, and the only outside calls are your own gh calls to GitHub. Daemon only listens on localhost, and every action shows its command before it runs. MCP server is opt-in, and sending to other sessions is off by default; if you turn it on, every send is a preview plus a separate confirm you approve. Details are in the README's #safety-and-privacy, and the MCP threat model (including what it doesn't protect against) is in docs/MCP.md. To report a security issue privately (or contributions to improve), see SECURITY.md. with mock data, nothing to install at https://shufler.app at https://github.com/jgrichardson/shufler (MIT)
原文—
Coding agent spots your AI practices and the ones you're missing
Frontier models improve at a pace faster than the pace at which most humans are able to keep up. Not all of us are obsessed with pushing the boundaries of our agents and trying new ways to work with them. I often spend 12+ hours a day working with an LLM, and still, when I talk to people, I hear of ways of working I hadn’t thought of. What if we could speed up that pace of learning? What if everybody shared their practices of how they are making the most of AI, and if we could learn from each other? What if AI could make that sharing happen? Could everyone catch up with AI then? I know, it sounds dreamy, but my therapist says illusions are necessary—until some of them come true. I built a tool that teaches your local agent to scan your sessions, spot the best practices you already use, and guide you towards the ones you haven’t discovered yet. Only names and counts reach the server. Your sessions never do. The product has a self-consolidating catalogue of practices, MCP servers, plugins and skills. Here’s my stack—scan (locally, all session data remains private). https://app.userrank.ai/@milstan To spice things up and give a reason to rescan, I added rankings inspired by Foursquare (who here is old enough to remember Foursquare when it was gamified?). You can rank in the top 10 of a city, or of a field like Product. Claude Code, 10–20 minutes on your own plan: claude plugin marketplace add https://app.userrank.ai/plugins/marketplace.json claude plugin install userrank@userrank claude "Get my UserRank" Codex, or a quick scan of your Claude or ChatGPT chats: https://app.userrank.ai/start
原文—
FinProof
Agent Siege: 6 agentic attacks against a banking AI — 60 seconds, no account. 11,474 adversarial prompts across 13 BFSI categories: • Indirect prompt injection via support ticket • MCP tool poisoning (metadata hijack) • RAG/memory poisoning • KYC bypass, PII exfiltration • Multi-hop traversal Mapped to RBI, SEBI, EU AI Act, SR 11-7, DPDP Act 2023. T1 public on HuggingFace. T4 withheld — no training on the test set.
原文1
Cognitum Metaharness
🛠️ The meta-harness for AI agents — scaffold your own focused, branded agent harness with its own npx CLI, MCP server, memory, learning loop, and witness-signed releases. Works with Claude Code, Co...
原文3
TryMermaid
TryMermaid is a visual Mermaid editor. Drag shapes on a canvas or write code, and both stay in sync. Supports flowcharts, sequence, ER, Gantt and 14 more diagram types. Export PNG/SVG, share links, embed anywhere, or connect AI agents via MCP. Free.
原文1
MailDignity
MailDignity helps teams catch generic, presumptuous, pushy, and artificial language in outgoing emails. Available as API and MCP integration, it audits and improves cold email outreach before messages reach prospects.
原文1
Railse：一个人同时开几个 AI 编程 agent，审批推到手机上点一下
最近几个月我一直同时开着 Claude Code 、Codex 和 Gemini CLI 写代码。agent 本身跑得挺好，烦的是它每次要跑命令、改文件都停下来等我按「允许」。三个终端轮流翻，经常翻到的时候它已经等了十分钟；出去吃个饭，回来每个都停在第一条 git push 上。开 yolo 全放行又不敢，一句 rm -rf 出去就没有后悔药。 所以我写了 Railse：每条审批先过一遍规则，规则定不了的推到手机，我在手机上点一下，agent 接着跑。 它怎么做： - 三家 CLI 的权限钩子都接到本机一个守护进程，规则只写一套。 - 每条请求先定风险档：只读和改项目里的文件直接放行； curl 、npm install 这类联网的，git push 、deploy 这类对外的，rm -rf 、git reset --hard 这类不可逆的，都问。 - 定不了的推到手机，Web Push 通知，点开批准或拒绝。批准时勾「记住」，当场存成规则，下次同类的不问。 - 对外和不可逆的永远不能被「记住」放行，只能自己手写规则。 - Railse 自己挂了不卡 agent：守护进程没起、hub 连不上超过 30 秒，CLI 弹自己的审批框，最坏就是回到没装的样子。 hub 自己部署，一个 Go 二进制，也有 Fly.io 的现成配置；手机网页由 hub 托管，不用另外部署。命令里认得出的 token 、密码在本机就换成 再上报。 现状：v0.1.2 ，我自己在用。已知的限制：Gemini CLI 的钩子只能拒绝不能放行； Codex 的钩子只在交互会话里生效； iPhone 要加到主屏幕才能收推送； hub 是单机 SQLite 。 开源，Apache 2.0： 一行安装：curl -fsSL | sh 想听两种反馈：你同时开几个 agent 的时候是怎么处理审批的；规则表有判错的（该放行的问了、该问的放行了）直接提 issue ，我会优先看。
—
Gear: Scikit-learn for RSI
Gear is an open-source framework for recursive self-improvement of AI agent harnesses (prompts, tools, workflows, and context management). Built toward a scikit-learn for RSI, its modular architecture lets developers implement, compare, and extend optimization algorithms. With Gear, GPT 5.6 Luna reaches GPT 6 Astra-level performance: 88.9% vs. Astra + Codex’s 84.1% on AutomationBench’s public Marketing tasks.
原文2
kensaku
Fast, local-first search for files, documents, code, and images—semantic search, OCR, hybrid ranking, CLI and MCP for AI agents. No cloud.
原文1
i / lninl
An MIT-licensed coding agent for your browser and terminal. Bring your own model API, edit files, preview HTML, export ZIPs, and extend the browser agent with Rust/WASM plugins. Free software; model API usage is billed separately.
原文1
Binary – hiring tool (ATS) for startups that is MCP friendly
Hi HN, I am Shikhar, founder of Binary. It's an applicant tracking system (ATS) for startups that is MCP friendly and comes with a free forever plan. It can do a lot of things - creating hiring forms, career pages, screening candidates, tracking the hiring funnel, scheduling interviews and more. I decided to build it because old-school ATS tools felt very slow, outdated and expensive. In fact, many of them don't even allow using the tool without speaking to sales. We not just have a self-serve, but the AI even sets up some sample data so that you can play around instantly. We've been heads down working with a lot of startups who all wanted a more AI-first ATS, and seeing some early success. This is my first time posting on HN. If there are any founders/recruiters, I'd love to get your honest feedback or thoughts on the tool.
原文—
ALLR
ALLR compares motorcycle gear prices across retailers and ranks them by landed cost: item price plus shipping, import duty and brokerage, shown in your currency. Built for Canadian riders, also covers the US, EU and UK. Free, no account. Includes a free import-duty calculator, 40+ buying guides, an AI-agent (MCP) server and a ChatGPT app.
原文2
Control the browser directly from TheGitAI Terminal
I am the owner and founder of TheGitAI Coding Agent. I built this Chrome extension to let the agent control your existing browser directly from the terminal without interrupting the workflow. It doesn't require your login details or any secrets. You can control your browser now through agent without sharing any credentials through UI. Each step is taken right in front of you with controllable permissions. Try it for your most complicated work that requires browser access. Would love to hear feedback!
原文—
范围澄清练习：让AI扮演临时加需求的客户，自己练一轮回应
自荐一个 AI 辅助制作的中文开源练习，面向网页接单新手：原本说好做三页静态网站，客户又想“顺便”加会员、预约和付款，怎么先把范围与未知说清楚？ 项目： 免费版使用一个固定虚构案例。使用者自己回应，AI 再引用实际原句，指出一项已澄清的边界和一项仍需确认的问题，不替使用者回答，不给职业或收入保证。 两种免费体验 不下载：复制 README 的单轮提示到自己已有的 AI 对话中，先回答一次；这种方式不提供本地保存与恢复。 需要保存：下载免费的本地 Skill，用具备文件读写和 Python 3.9+ 的 Agent 主持；自备模型权限和额度。 免费仓库按 MIT 开源。仓库另说明同作者的独立付费版本，本次只邀请体验免费版，不要求购买。普通 AI 也能做基础沟通练习，固定流程是否有额外帮助仍需要真实反馈。 想听到的反馈 你实际试了哪种方式？哪里最费劲？自己回应后，哪一句反馈有帮助或没有帮助？还没试也欢迎说明原因。可用 README 的体验表单或邮箱反馈，也可以在这里直接讨论。 请只使用本项目的虚构案例，不提供真实客户、订单或个人资料。当前只有本地技术核查，没有真人学习效果或成交证明；此次希望先找到愿意亲自试一轮的人。
0
BrandZap Kit
Each Kit connects your custom fonts, colors & tokens, UI components, graphic assets & icons, and voice to tools like ChatGPT, Claude, and Cursor through MCP. It provides a shared foundation for creating marketing content and product interfaces that reflect your brand, without repeated prompting and dragging PDFs. At its core, is a "thinking layer" that works with your LLM of choice to provide all of guidance and insight needed to keep everything you and your team makes, on-brand.
原文2
Diali.ai
Text your AI a job on WhatsApp, Telegram, Slack or Discord, and it gets done. Diali runs an always-on OpenClaw or Hermes assistant for you: it browses the web, can drive its own computer while you watch live, remembers you and checks in on its own. Pick from 17 Claude and GPT models, bring your own key, or use your ChatGPT plan. 75,000+ skills, 300+ MCP servers, 250+ CLI tools, each assistant in its own sandbox. Ready in about a minute, no servers to run. Try Starter free for 7 days.
原文1
FrameLoom：把文档做成风格统一、可逐幕修改的图解视频
阮老师好，自荐一个我维护的开源项目 FrameLoom。 它是面向 Codex、Claude Code 等 Coding Agent 的视频制作 Skill。输入产品说明、文章或讲稿，可附真实截图，由 Agent 编写讲稿和分镜，再通过 Remotion 在本机渲染，适合产品介绍、功能更新和教程视频。 项目把风格、镜头、配音字幕和检查整理成一套可复用的制作流程，主要提供这些便利： 先看样片，再选风格： 提供实际渲染样片，选好画法和字体后，后续视频可以沿用同一套选择，保持系列外观一致。 基于原项目逐幕修改： 保留讲稿、分镜和素材，可以让 Agent 修改某一幕的文字、替换截图或调整讲解顺序。 制作环节已经衔接： 支持 TTS 或外部旁白，按实际音频时长调整镜头，检查文字裁切、遮挡和时长；也能导出干净画面底片供后期剪辑。成片仍需人工复核。 通用 Agent 也能自行搭建这些能力，FrameLoom 提供现成的实现与流程，方便持续制作和修订视频。
2
Kodl Agents
Kodl now brings four specialized AI agents into your team workspace. Delegate a task to the Kodl Agent, Writer, Planner or Coordinator, review contextual proposals and approve changes before they are applied. Track runs, sub-agents and estimated costs in one view. Alongside Kanban, sprints, chat and docs, Kodl now includes desktop apps, offline support and sprint forecasts. Built for teams that want AI assistance while staying in control.
原文1
App Store Screenshots
A free, open-source skill for Claude Code, Codex, and Cursor. Give your agent your app captures and a short brief, and it builds a local screenshot editor in your project. Choose from 18 styles, compare looks in Style Lab, edit the copy, and export every size for iPhone, iPad, Mac, and Android, in multiple languages. No screenshot subscription. It runs locally; your usual agent costs apply. Install: npx skills add ParthJadhav/app-store-screenshots
原文100
runspace
runspace runs dependencies, build scripts and AI agents' commands on a disposable GitHub codespace rather than on your machine. For devs who want builds and tests off their laptop. For people who run coding agents and want every agent command in a sandbox. Has a CLI and an MCP server with the same commands, returns the work as a patch you can check before applying, and requires only a single binary and a GitHub token.
原文1
Specbook
Much more than a Swagger UI alternative. A live documentation portal that opens any OpenAPI or Swagger spec and runs it against mocks or the real API. Compose flows in TypeScript with your AI agent, even with third-party integrations like payment widgets, and run them in a QA portal, in CI and as production monitors.
原文2
Interview your analysts instead of auto-generating a semantic layer
I’ve spent the last year building context for AI analysts in the mid-market. I’ve open sourced a library from my learnings, along the same principles of the Matt Pocock skills. Principles: * Claude / Codex / Cursor etc are great data agents - you just need context to enable them to answer business questions correctly. * Much of the important context lives in analysts heads' * Context must be maintainable (short, human readable - markdown and yaml) * Added context should be load bearing, not repeat of information that is available in another source of truth - and evaluations prove the value of the context * Companies should own their own context, and must be shareable with any model * Context should be version controlled (and github works as a context store). What the library does: * Performs an interview with your analysts (context-interview skill), one domain at a time, similar to what I have done as a consultant. The interview is informed by your query logs. * The AI develops evaluation questions at the same time as building the context - and at the end of each domain interview asks you to run an evaluation with and without the context. The evaluation will actually check the results against your dashboards if you allow it. We have shown increases in accuracy from 87% to 95% by adding this context in our imaginary open-source company, Shorelane. * Provides a set of skills to use the context that follow best practices for self service analytics: * Write a plain english plan for the business user to approve * Execute the plan * Verify that the execution followed the plan, and provide a simple uncertainty score. * Enable a adversarial review of the answer if their are doubts. What you get: Context written in markdown and yaml after you finish the interview. We use a single github repo, divided into domains. The interview can be shared and restarted by checking in/out of github. A set of skills to share with your business team Optional: We offer a fast MCP server to serve the Github-based context to your team.
原文10
Turnbook
Turnbook keeps every conversation with your AI agents word for word on your own PC and shares it across all of them over MCP. Saving costs zero model tokens; a question gets back ~3k tokens of evidence. 88.8% on LongMemEval. Free and open source.
原文1
asd-ste100-skill
用简化英语改写 AI 指令。该项目是基于航空航天业的简化技术英语规范，能把含糊、冗长的英文改写成不会被误读的句子，然后输入给 AI 智能体。它会逐句检查被动语态、复杂时态、一句多指令、长句和营销形容词等问题，改写时不丢失原文的任何事实和条件。
5750
Our Silly Games
Our Silly Games is a free home for browser games made with Claude Code or Codex. Install the plugin, say "make me a game about...", and publish. We build it, review it and host it, with multiplayer, leaderboards, saves and phone controllers built in.
原文68
artemis
谷歌开源的 AI 手机自动化工具。这是一款 AI 驱动的 Android 自动化工具，只需用自然语言描述任务，AI 就会自动操作手机或模拟器，完成跨应用的测试流程和日常任务。它内置 MCP 服务，能让 Claude Code、Codex 等 AI 编程助手直接操控手机并收集日志和截图，支持 Gemini、Claude、GPT-4o 等多模态模型。
—
Scout by Seedflows.ai
Scout researches business opportunities that fit your skills, budget and time. AI agents check demand and competition, provide sources and review each other's findings. Turn a promising opportunity into a project plan. Pay for usage, with $2 to start.
原文1
—
IdeaProof.io - Startup Idea Validator
IdeaProof is no longer just a validator - it’s the Founder OS. We turned your raw idea into an end-to-end launch pipeline: • Real demand intelligence: Stress-tested against 1,000+ startup post-mortems. • Sequoia-style pitch decks: Auto-built from your market data with 1-click PDF. • Complete visual identity: Ready-to-use color schemes, typography, and assets. • Native MCP Server: Run validations, pull failure stats & stress-test ideas straight inside Claude, Cursor, or your terminal.
原文1
Synapse
Synapse is a free, open-source desktop AI workspace that brings local Ollama models and cloud AI providers together. Chat with persistent memory, use MCP tools and custom skills, manage context automatically, and build interactive experiences inside conversations.
原文1
Intelligent for Keynote, Pages & Numbers
Apple is bringing intelligent assistants to Keynote, Pages, and Numbers. Create presentations, write and edit documents, analyze spreadsheets, and build charts using natural-language instructions, right inside the apps. With new MCP support, compatible external AI agents can also read, edit, and work with your files directly. No more copying content between your AI assistant and Apple's productivity tools. Coming in November as part of Apple Creator Studio.
原文2
opendaw-mcp
MCP server for agent-native control of openDAW: create tracks and notes, route effects, mix, and render audio in headless Chromium. Written in Python. Open source, Apache-2.0. Source, docs and issue tracker:
原文1
MentionAgent
MentionAgent runs link building outreach for you. It finds blogs in your niche that could link to you, finds the author's email, writes a personal pitch for each post and sends it from a warmed-up inbox it sets up, so your own domain is never touched. Replies land in one inbox. Agencies run it per client site. Plans start at $99/mo, priced per sending inbox with no per-seat fees. Works from Claude or Cursor through its MCP server.
原文3
Listbulb
Building a product is hard. Getting it noticed is harder. ListBulb helps indie makers launch products, discover new tools, earn community Bulbs, and build lasting visibility with permanent listings and dofollow backlinks. You can also launch sooner with QuickPass or get featured for extra visibility. Building something for developers or AI agents? Explore our REST API and MCP server to bring product discovery into your own tools and workflows. Let's help good products get discovered. 💡
原文2
LIVEAIHELP
LIVEAIHELP connects you — or your AI agent — to a vetted human expert in minutes, over live video or chat, when AI hits a wall: a CAPTCHA, a login, a judgment call. • On-demand matching: no job posts, no bidding, no waiting • $1/min, pay only for what you use. Bundles from $10 • Bot API + MCP server: your agent can call hire_human_expert when it's stuck • Vetted experts: AI exam + background check • In-page remote control with your permission
原文1
Krill
Run Claude Code, Codex, Cursor CLI, OpenCode or any shell side by side, each in its own Git worktree, with a queue of agents waiting on you and diff review. A separate daemon holds the sessions, so agents keep running when you quit, crash or update the app. Rust desktop app, no Electron or web view. Oct 28 benchmark (Krill 0.2.20 vs Orca 1.4.242, 10 terminals, screen on): 1/7 the memory, idle CPU and install size. Free, Apache-2.0. Apple Silicon Mac; Windows preview.
原文1
Local, Tracked, Versioned, Legal Research Databases
This is a project I've been working on for about 8 months now. I worked in legal tech for 3 years and found that one of the biggest hurdles for AI agents and lawyers alike when doing legal research is tracking changes, and correctly cross referencing those changes between different forms of legal documents. W2D builds a SQLite dataset from public sources, and offers agent skills for answering questions about that dataset. The dataset holds multiple revisions of the U.S. Code at each release point that you add. It holds the public laws that amend the Code, and it links each amendment to the change that it made. It holds court opinions from CourtListener, by page and paragraph, and it links each U.S.C. citation to the provision that the court cited. It can also hold House roll calls. This is similar in spirit to "git for Law" but includes a lot more utilities than diffs to make it an effective tool for research.
原文—
OpenAny
OpenAny is a marketplace for AI agents, bringing models, APIs, skills, and more into one place. Access the capabilities your agent needs with one API key and one shared balance, without managing separate accounts and top-ups for every service. From model inference to data retrieval and task-specific workflows, OpenAny simplifies how you find, connect to, and pay for services. Agents can also sign in using a wallet or email account you authorize.
原文1
bytemail
bytemail is a native Mac email app built in Swift. Gmail, iCloud, Outlook, Yahoo, Fastmail and any IMAP account in one fast window. New mail and sign-in codes grow out of your MacBook's notch, Undo Send flies a paper plane through it, and desktop widgets show what's unread. Tracking pixels are blocked, look-alike senders flagged, and your mail never touches our servers. Snooze, rules, one-click unsubscribe, and an MCP server for Claude and other AI apps. iPhone beta included. 7-day free trial.
原文225
AutoFire
Since our first launch, AutoFire has grown beyond dealership websites. Shoppers can describe the car they want, search live inventory, and ask questions through an AI website assistant. Dealers manage inventory, leads, conversations, and test-drive requests in one dashboard. LotIQ provides daily briefings and follow-up suggestions. Connect Claude or ChatGPT through MCP to read dealership inventory, leads, and test-drive requests, with access limited to the dealership and permissions you approve.
原文2
In-repo delivery board: stages, handoffs, soft WIP, weekly report, Timeline
Project URL Category JS Project Title In-repo delivery board: stages, handoffs, soft WIP, weekly report, Timeline Project Open-source delivery board that stores stages, owners, handoffs, and soft WIP limits in a git-tracked file. Includes a weekly Markdown client report and a local kanban plus calendar Timeline UI. Runs standalone via npx or as a DeepSeek Harness plugin. Aimed at keeping delivery context next to the code rather than only in SaaS boards. Highlights Delivery state lives in the (delivery.json) and is reviewable in PRs Automatic handoff audit when cards change stage; soft WIP limits warn without blocking Weekly Markdown report (shipped / handoffs / risks) plus local editable board and calendar Timeline Standalone CLI: npx jerryxff26-alt/dsh-delivery-board — no SaaS account required Example Code bash npx jerryxff26-alt/dsh-delivery-board init '{"customer":"ACME","template":"governance"}' npx jerryxff26-alt/dsh-delivery-board card '{"title":"API integration","stage":"analyze","owner":"Ada"}' npx jerryxff26-alt/dsh-delivery-board open npx jerryxff26-alt/dsh-delivery-board weekly or Videos GIF in
原文4
AutoArb
Give Codex, Claude Code, or any MCP client current vehicle listings, a maximum acquisition price, and the evidence behind the number.
原文2
Nano Metrics
Nano Metrics adds a live dashboard to your VS Code sidebar: Claude Code status, tool calls, and task progress; token usage and estimated costs from local transcripts; edited files and recent activity; CPU, memory, disk, GPU, per-core load, and temperature; optional monitoring for 30+ services like Docker, Postgres, Redis, Kafka, and Ollama; process-level usage; and a theme-aware status bar. AI coding activity, costs, and system metrics, all in one place.
原文1
Aigent-OS, a Claude Code harness that clears and resumes
I run my business on four Claude Code agents: Titus, Pheme, Davinci and Metis. They write my newsletter, The AIgent (about 9k subscribers), every day through a drafting and proofreading engine that Pheme checks before scheduling. They also build AI courses, run a pipeline that turns the day's featured into a talking-head video of my avatar for Instagram, set up ad deals, and run Meta ads and email funnels with me. Nothing closes until I sign off. Metis posts jobs on a job board and the others claim them. The hard part wasn't any single task. It was keeping the agents coherent over weeks. Compaction kept dropping the small decisions, and then an agent would redo work I'd already ruled out. aigent-os is my fix. At a context threshold (80% by default) the agent writes a capsule, a checkpoint file in git, and has to reply with one exact line. Only then does the runner send /clear, wait for a SessionStart receipt proving a fresh session started, and resume. One ack buys at most one clear. If anything is ambiguous it stops at HOLD:clear-stalled-after-ack and doesn't retry. On resume the capsule is reference only, live state wins, and anything the writer couldn't capture is written as Unknown. First run interviews you with three plain questions and makes a first plan. Each agent gets its own memory vault. Limits, honestly: the public evidence is 11 of 11 cycles on one Windows seat. The runner can still miss a permission dialog (#104), a seat launched from inside another Claude Code session can lose its transcript (#98), and scrolling the seat window sends stray input (#99). v1.1.0 shipped on Sunday. Try bash install.sh --dry-run, it writes nothing. I'd like to hear where the loop could still double clear or strand a session.
原文—
Similarity, a spiritual successor to 2000s-era OkCupid
This is a side project I've been working on on-and-off (mostly off) since the 1990s. I guess you would call it an experimental "dating site" though it's designed to be fun/interesting to people who aren't single (like me). You don't need to sign in to see how it works, so try it out. There is no swiping. The original inspiration was a 90s-era website called moviecritic.com that let you rate movies and get movie suggestions via collaborative filtering. I thought "hey, what about just matching me to the _people_ with the same taste instead?" And why just movies? The closest thing to this was 2000s-era OkCupid, before it was Tinderized. I want to capture its appeal, while making something novel. Similarity is not vibe coded, though I lean pretty heavily on codex (mostly the $20/mo plan - not a lot of tokens burnt here). The stack is java, guice/gwizard, elastic beanstalk, postgres (neon), react, jotai, and few cloudflare doodads. The match engine is a custom in-memory structure; getting the LLM to load test and optimize it was really a wow moment. Everything is PaaS and designed to scale by itself, cheaply, hands-off. I'm working on some ideas to help with the chicken/egg problem but for now I have low expectations. It's cheap to run and I have a day job. Would love feedback!
原文—
Aident Loadout：让多个 AI Agent 共用一套应用连接和 Skills
同时用好几个 Agent 时，每个都要单独配 MCP、API Key 和授权，很麻烦。Loadout 提供一个远程 MCP 服务器（ npm CLI（@aident-ai/cli）和一个开源 Skill，Agent 按需求搜索应用动作、读取实时 schema 再执行；另外内置 400+ 个现成的 Skills（调研、内容、运营等）。OAuth 在浏览器里授权，API Key 保存在 Aident Vault 里不交给模型，每次调用都会记录在审计日志里，并显示花费。可以免费按量付费开始使用。 我是 Aident 团队成员，本帖为自荐。
7
Muha：本地Harness CLI的统一接入层
Muha 是一个 MIT 开源的 TypeScript SDK，用于把已有Coding Harness接进脚本和应用。它统一处理启动、会话、事件和交互，让调用方用代码连接不同 Agent 的任务。
2
Crulon – Business phone where AI does the work, humans make the calls
Hi HN, I’m Rayhan. Crulon is a business phone system where an AI agent answers the line and your team works alongside it. You can talk with our agent right in your browser, no sign-up: https://crulon.ai/#talk Why I built it: I’ve spent eight years as a software and network engineer, mostly building tools for internet providers, and kept watching small support teams lose customers in the seams. Calls missed after hours. Calls passed from a bot to a person who had to start over. Most voice AI products are a bot sitting on someone else’s phone line, so a handoff means forwarding to another number and losing the context. We built the phone system and the agent as one thing. What that gets you: The handoff happens inside the same system. When something isn’t the AI’s job, it asks the caller, then rings your team (in the browser or on a desk phone) with a note on why they’re calling. Live calls: anyone with permission can listen to an AI call (no one on the call can hear them) and take it over. It’s the same call, with no transfer and no hold music; the AI just leaves. Tools: the agent can call any API you define in an Action builder (look up the caller, open a ticket), and it checks with the caller before changing anything. Outbound campaigns: the agent works through a call list for reminders, renewals and follow-ups, and puts interested people through to your team. SIP desk phones, and bring-your-own-carrier so you can keep your numbers. If you run out of AI minutes, calls follow your fallback route (ring the team, voicemail) instead of dropping, and the refused call isn’t billed. How it’s built, briefly: calls come in over SIP into Call Flows (menus, queues, schedules), and your team answers in the browser or on SIP desk phones. Each agent runs on one of two engines. Classic transcribes, thinks, then speaks, so you get exact wording and English, Spanish and French. Live is a full-duplex speech model that listens while it talks and hands tool work to a second model. One fun problem with that split was teaching the agent when to hang up, which I wrote up here: https://crulon.ai/blog/teaching-a-voice-agent-to-hang-up It isn’t a CRM; it works alongside yours. Built-in Apps connect it to systems like billing and network tools, and anything with a web API plugs in through the Action builder. Pricing is per seat ($24, $49 or $89 a month) with a pool of AI minutes; human calls are included. If you sign up this month, HN30 takes 30% off the first 3 months. I’ll be here all day. Try the agent, try to trip it up, and tell me what you’d want from a phone system like this.
原文—
Offer来敲门：中文本地 AI 求职工作台，串起岗位简历、投递跟进与模拟面试
求职时，岗位描述、不同版本的简历、投递进度和面试反馈经常散落在多个地方。Offer来敲门把这些步骤放进一个本地运行的中文工作台，适合已经使用 AI CLI、希望系统整理求职过程的朋友。 可以从一个目标岗位开始：手动添加职位，结合自己的真实经历生成岗位专属简历和沟通话术，核对后导出 Word、PDF 或 Markdown；自行投递后记录进度，再围绕该岗位生成面试题，进行文字或语音练习并查看逐题反馈。 也支持多个招聘来源的职位发现与匹配分析。需要登录的平台由用户自行登录，来源可用性受平台页面与访问限制影响，工具不自动投递。 运行需要 Node.js 24+，以及已登录、有可用额度的 Codex 或 CodeBuddy CLI；Claude Code 已实现适配，但尚未完成真实账号流程验证。项目采用 Apache-2.0 许可证，模型使用可能产生费用。 档案和求职记录保存在本机；AI 处理仍会将经历、技能等任务内容发送给所选 CLI 配置的服务，不是完全离线工具。 欢迎试用后反馈：哪一步最有用，在哪一步遇到阻碍？反馈请附操作系统、引擎及版本、操作步骤和脱敏错误信息，无需上传真实简历。
3
Sectr – Serverless backend platform for easy deployment of AI agents
Hey HN, I've been building Sectr, a backend-as-a-service for making it easy(ier) to develop and deploy AI agents. Deploying agents requires non-trivial systems design work. Consider a support agent: how do you stream messages to your frontend, how/where do you persist messages for frontend fetching later (or for auditing), and when the agent wants to issue a refund, do you spin-down the compute infra while it's waiting on human approval? Sectr aims to be the serverless abstraction and platform to provide these systems for you. No sign-up needed to try using the dev server: curl -fsSL https://sectr.dev/install.sh | sh # Install CLI sectr init my-agent && cd my-agent # Starter code sectr dev # Dev server, with UI to test chatting Sectr provides: - Streaming of every token, tool call and event, and persistence so they can be loaded fresh later. - Scaling up, and down to zero when the agent needs to wait, and resumption right to the suspended agent's state. - Human approved tools automatically use Sectr's suspension and resumption capabilities, no code needed from you. - AG-UI endpoint for any deployed agent so you can trivially hook up your frontend - Local dev server experience, so you can rapidly debug and iterate on your agent's behavior At the core, you can deploy any container image to Sectr as long as it implements a certain shaped HTTP endpoint. In practice, currently Sectr provides adapters to let you deploy your existing OpenAI Agents SDK, LangGraph, or Pydantic AI agents, with only 2 new lines of code: app = AgentApp(name="support") app.add_route("/chat", openai_agents(agent)) Certainly the hardest part of building Sectr was getting the resumption logic to be correct and invisible, across the Program State x Model Provider x Agent Framework matrix. For example, an earlier bug was when a model replied with text plus a tool call, Sectr would resume that as two separate messages, instead of one message with a text and tool part. This works, but breaks prompt caching (and might subtly change model behavior). Sectr is free during the beta (and can be tried locally without sign-up); I mainly want feedback, especially on dev & deployment experiences, any bugs, and what features would excite you. Introductory video: https://www.youtube.com/watch?v=H-y3wynkKc0 Docs: https://docs.sectr.dev
原文—
Castellan – A real-time strategy game for agents
In an RTS game, making decisions quickly matters as much as making good ones. Looking for the balance between cost, speed and intelligence in agents, I started wondering what an agent playing an RTS would look like. So I built a simple RTS, and I keep making it more complex. - The game keeps running while the agent thinks - The agent can upload a JavaScript script for each unit type - 300+ game replays - Connects through MCP or CLI. Works with Claude Code, Codex, Cursor CLI, Antigravity CLI, OpenCode, OpenClaw, DeepSeek Harness and MiniMax Code - 2 to 6 players, FFA and co-op, open and walled maps - 4 units, the King and 4 buildings - Open source, written in TypeScript
原文—
Coday：让 Devin、Cursor 等编辑器使用你自己的模型
Coday 是一个 Windows 桌面工具：在本机接管 Devin、Cursor、Windsurf、Kiro 以及 VS Code（GitHub Copilot）的 AI 请求，转换协议后转发到你自己选择的 OpenAI 兼容或 Claude 接口。编辑器不用手改 baseurl，聊天、补全、Agent 和工具调用照常使用。 主要功能： Devin 全家桶：快速上下文复用本地 Devin 登录；Devin Tab 补全走官方 GetTab Supercomplete；本机 Devin CLI / Devin Local / ACP 的系统提示词改写与回滚；Devin 界面汉化与还原。 自选模型与 API 源：粘贴接口地址和 Key，同步上游模型列表；支持多源故障转移，当前源限流或不可用时自动切换。 跨 IDE 项目规则：读取 .windsurfrules、.cursor/rules、.kiro/steering、CLAUDE.md 等规则文件。 Remote-SSH：通过反向隧道把 Linux 远端的 AI 流量接回本机，路由和模型注入继续生效。 可观测：内置请求控制台、用量统计、证书与拦截诊断。 界面支持中英文，仓库提供 Windows 安装包，源码暂不开放。
54
Agent-Ready Shop
A WooCommerce plugin that lets AI assistants like Claude and ChatGPT search your products, check live stock and prices, and send the shopper a pay link on your own checkout. Browse-only by default. No marketplace cut. Orders land in WooCommerce as usual.
原文2
On Record Verify
On Record Verify lets any AI assistant or platform check whether a photo or video is a genuine On Record capture: when and where it was taken, and whether it has been changed since. Paste the code on the picture into Claude or any MCP-capable assistant and it answers from the record, not a guess. Every capture is sealed at the moment it is taken by On Record's Authenticity Chain and anchored into Bitcoin. The proof is public, so nobody has to take our word for it. Free, read-only, no sign-in.
原文1
Thinker
Thinker cuts your coding agents’ token spend by ~20%. Thinker is an open-source proxy that sits between your agents and the model. It strips context your agent no longer needs, caches knowledge so future agents don’t have to rediscover it, and keeps your instruction files lean and current. Across my first dozen users, it’s averaged 20% fewer tokens per completed task, with no drop in success rate. For teams, it also shows where token spend is going, by team, repo, and agent.
原文1
Limma – MCP for total recall and precision in data retrieval for agents
Hello, my name is Berend Graham. My cofounder Shiv Oberai and I have been working on Limma for a year now with the goal of building a proactive solution to the reliability issues that plague practical AI and agentic use in companies with RAG. Limma is a single MCP that exists for the sole purpose of ensuring that the right information is retrieved with every query and no information is skipped and or missed. This means total precision and recall. We've found that this approach leads to the lowest rate of hallucination vs. solely targeting agent action verification which has made tremendous strides in capability. We do this by building a map and structuring all data-points within unstructured data as well as federating structured data for addressability so that agents don't need to search through chunks directly and instead only get the right information . We wanted this to be extremely easy to use so that anyone can set it up and try it within minutes so that the precision issues (especially with large quantities of data) that LLMs struggle with can finally be remedied. We use self-hosted models (primarily Qwen models as of current) so that the system can be deployed on one's own server if needed, and also to optimize for speed. We have a free monthly version on our website if anyone would like to try it out and provide feedback!
原文—
TeapotRouter
按用量计费的 AI 模型网关，一个 Key 双协议通用：Claude Code 设置 ANTHROPIC_BASE_URL、OpenAI SDK 改 base_url 即可接入，价格公开、每笔请求成本可查，支持组织内多发 Key 与用量审计、多上游自动故障转移；名字与 logo 源自 HTTP 彩蛋 418 I'm a teapot
—
GoZapp
GoZapp is a job search app for France. Its free, unlimited AI Agent reads your profile (skills, experience, location, salary), scores every new job offer out of 100 and sends you the best matches. GoZapp also writes your CV and cover letters with AI and checks your CV against applicant tracking systems (ATS). On the web, each tool costs 1 euro with no account. Available on iOS, Android and the web, in French.
原文1
Pixtorium
An approachable, AI-native image editor: layers, real photo adjustments, retouching, artistic effects you can paint with, on-device AI, asking for edits in plain words, and an MCP server for AI agents. Not a PhotoShop clone! Free in a browser, no account needed. Use brush with every effect!
原文1
Token Kingdom
Token Kingdom turns your Claude Code usage into a cozy 3D village. Every token you spend becomes gold, wood and stone. Build, fish, cook, defend against raids and visit your friends' kingdoms. One plugin install. Free.
原文2
Google Apps Script Copilot
Since our first launch, GS Copilot has grown into an AI coding agent inside the Apps Script editor. New: Agent Mode for multi-file changes, Plan Mode, "Fix with GS Copilot" to debug errors in one click, inline suggestions, and MCP Tools on Pro and Lifetime. You approve anything that sends, deletes or touches a secret. Launch offers until Oct 31: code PRODUCTHUNT takes $30 off Lifetime + BYOK ($99 one-time), and code PHSUB takes 25% off monthly Plus or Pro for your first 3 months.
原文1
Doska
Doska is a whiteboard + docs where diagrams are code. Write Mermaid or D2, or drop terraform show -json, and get editable shapes synced with the code. Embed live diagrams in docs, let Claude edit via MCP. Free sandbox, no sign-up.
原文2
Raq.com
Raq.com started as an internal tool builder. It's now a toolkit for the AI agent you already use. Connect Claude, ChatGPT or any MCP agent once and it works across 75+ business tools, from CRM and task boards to quotes with e-sign and card deposits, forms, automations and AI image and video. Everything it makes is saved in one workspace your team shares, and you can build and host your own custom tools too.
原文2
Bifurc
Free desktop app for API work on localhost: *.localhost proxying with capture, a client for REST/GraphQL/gRPC/SOAP/WebSocket/MCP & more, proxy-level mocks, git-synced workspaces, visual flows, a local mail inbox, and an agent-ready CLI. Windows, macOS, Linux.
原文2
SecondBrain
New since our June launch: SecondBrain is rebuilt around your voice, and it now runs in your browser too. Tap and talk: it writes a clean note in your words, files it, and turns dates into events and reminders. No time given? It asks. Share links, PDFs and photos from any app, then watch it work live. Ask anything and the answer streams from your notes. Pro adds one shared memory for Claude, ChatGPT, Gemini and Cursor over MCP. iPhone, Android and web. Free: 10 notes and 3 AI requests a day.
原文4
Scenario for Blender
Generate 3D, video, PBR materials, audio and images without leaving Blender. Scenario for Blender is an open-source extension (GPL-3.0, 800+ GitHub stars) that imports each result straight into your open scene: a photo becomes a textured mesh at your 3D cursor, a move you describe plays on your rigged character, one line of text blocks out a room, and your camera move comes back as a photoreal shot. One click connects Claude Code, Codex or Cursor too.
原文3
FYN
FYN is personal finance with one ledger for you and your AI tools. Link banks once on Pro. FYN categorizes transactions, keeps budgets and bills current, and turns receipts into line items. Ask FYN in-app, or connect Claude, ChatGPT, Grok Bot or any MCP client for a read-only view. FYN never moves money. Free forever: receipts, tap-to-pay and email ingest, Ask FYN. Pro: Plaid bank sync, higher limits. Households share one workspace. iOS, Android, web.
原文1
Velocity
Velocity tracks live aircraft (ADS-B) and ships (AIS) from public feeds with no API keys, records every position it sees to your own disk, and lets you drag a scrubber to rewind the map. Aircraft also replay from public archives back to 2024. Area alerts go to Discord, a webhook or ntfy. Every contact shows which feed reported it and how old the fix is. It also runs as an MCP server, so an agent can ask what is flying somewhere right now. docker compose up, AGPL-3.0.
原文1
Imaginou for AI agents
Create AI product videos from photos or a public product link. Connect Imaginou to an assistant over MCP and OAuth, review the scene storyboard before generation, then inspect and download the finished video. Choose showroom, presenter or comedic campaign workflows in four languages. Three free trial orders are available once per verified account, followed by one-time order packs.
原文2
Whenever (by Wix)
Describe the job in plain words and Whenever's AI builds the workflow, asking about your business rules, not technical settings. AI writes it once, then code runs it. Add AI steps or agent steps only where the job needs judgment. 1,400+ integrations plus any API. Runs on schedules, webhooks, app events or on demand; every run and step logged. Build in the app or from Claude Code, Cursor or Codex over MCP. Each workflow is a TypeScript file you can export. Free plan; Pro $39/mo, priced per run.
原文2
Scento Perfume API
Scento Perfume API is a remote MCP server and REST API. Paste into Claude, ChatGPT, Cursor or VS Code and twelve read-only tools appear. It covers 1,400+ designer and niche perfumes and 400+ raw materials, and needs no key or account. With the free browser extension, a small card on a Fragrantica perfume page shows what a 2, 5 or 8 ml sample costs, its rating and its leading notes.
原文3
—
AgentBreakWatch
AgentBreakWatch tracks release velocity, breaking changes, and MCP tools/list schema drift across 258 production AI Agent SDKs and Model Context Protocol servers on npm and PyPI. Paste your package.json, requirements.txt, pyproject.toml, ormcp.json to spot archived servers and breaking upgrades in 5 seconds, or query our free MCP server directly from Cursor and Claude Code.
原文2
Asonix
Asonix helps iOS app teams find keywords they can actually rank for, compare opportunities across 175 countries, and track rankings daily. See competitors' Apple Search Ads and Meta ads, monitor AI visibility, and connect to AI agents via MCP.
原文24
Filmit.io
Filmit.io is a suite of tools for editors and producers. • 50+ plugins, effects and transitions inside Premiere Pro, After Effects and DaVinci Resolve • Filmit MCP: connect Claude to Premiere, After Effects and Resolve to cut, add markers and build graphics • Filmit Studio online (beta): plan your story anywhere, review with clients, send it to Premiere, plus a production suite for estimates, schedules, call sheets, crew and releases One subscription, 7-day free trial.
原文1
Taxley
Taxley lets directors of small UK limited companies prepare and file their own Corporation Tax return and accounts online. Free to prepare; pay per filing, no subscription. Free MCP tools for AI assistants (Corporation Tax, deadline, penalty, director's loan and capital allowances calculators) at
原文1
OpenMontage Drama Studio：可自部署的 AI 短剧工坊，一句话灵感到成片
OpenMontage Drama Studio 是一个用 AI 做短剧和短视频的 Web 工坊。输入一句话灵感，按剧本、角色、场景、道具、分镜、配音、剪辑七个环节推进，每个环节的结果都能单独修改或重做。除短剧外，也内置热门短视频套路和电商带货广告模板，适合个人创作者和小团队自部署使用。 亮点 七个环节可逐步推进也可一键跑完：角色、场景、道具图会作为参考传给分镜出片，改一个角色或一个镜头不用从头来 爆款内容广场：抖音、小红书、TikTok、Reels、Shorts、B 站的结构化套路，可直接带入项目 广告片：内置电商带货转化模板（痛点 → 产品特写 → 卖点字幕 → 行动号召） 无限画布把项目流程显示为节点图；Skill 市场可安装分集剧本、分镜提示词等技能 文本模型自选 OpenAI 兼容接口（DeepSeek、OpenAI、Claude、Gemini、通义、豆包等）；出图 Seedream（也可接自己的 ComfyUI）、出视频 Seedance，二者均为需自备 key 的付费第三方 API React + FastAPI + PostgreSQL + Redis，Docker Compose 部署，AGPL-3.0 说明：项目最初基于 AGPL-3.0 的 calesthio/OpenMontage 代码起步，现已重构为独立产品，与原项目无隶属关系
4
WP AI Manager
WP AI Manager is an AI agent inside your WordPress admin. Ask in plain language and it edits Elementor and block editor pages, writes SEO, runs WooCommerce and 12 playbooks. Approve big changes, undo any. BYOK MCP for Claude and Cursor & more.
原文1
ars
An engine that learns your coding style from past projects, turns it into agent skills, and proves coding agents follow it. Claude Code today; Codex and other agents next.
原文1
AI Agent Approval Policy Generator
Create a JSON approval policy for risky AI agent actions. Pick action, environment, sensitivity, impact and reversibility. Runs locally with no account. Optional $19 pack adds editable policies, reviewer runbooks and integration checklists.
原文1
yovoice
yovoice is an open-source, local-first studio for turning words into expressive speech, cloned voices, full songs, and multi-track scores. Run everything on macOS, Windows, or Linux with no cloud API calls or per-character fees. Build stories and music, compare takes, edit timelines, export stems and MIDI, and connect AI agents through the built-in CLI, HTTP API, and MCP server.
原文1
Larder：本地优先的 AI Agent 资源管理器
同时使用 Claude Code、Codex、Cursor 等多个 Agent 后，Skills、MCP Server、CLI、指令文件和自定义命令往往散落在不同目录，很难确认本机到底装了什么。 Larder 会扫描这些本地资源，在一个 macOS 桌面应用里集中展示来源、Agent 可见性和更新状态，并支持带计划与快照的安全安装、更新、卸载。同时也可以浏览各 Agent 的本地会话和按需预览消息。 主要功能 Skills / MCP / CLI / 指令文件 / Commands 统一管理 Agent Inspector 解释 Agent 实际能看到什么 本地会话浏览、搜索和筛选 所有数据都保留在本机，不需要账号，不上传内容，也没有遥测。项目使用 MIT License，目前是 0.0.6 早期版本。
3
Archestack
Search and compare 322 open-source AI and vibe coding tools: chat UIs, model servers, RAG, agents, coding agents, Claude Code skills, starters. Each gets a score from public GitHub data, a review and its license in plain words. Nobody pays for a spot.
原文1
rederive：把代码库的设计决策变成问题，让你自己推导一遍
rederive 是一个面向 Claude Code、Codex、Cursor 等 25 种编程智能体的技能。它不直接讲解代码，而是读取代码库，找出塑造代码的设计决策（比如中间件为什么是一条链、状态为什么放在那里），把每个决策变成一个问题。你先自己推导设计，它再对照真实实现，指出你的答案和代码的不同，然后进入下一个决策。 适合学习 Express、Redis 等开源项目，或在面试前弄懂自己用 AI 写的项目。 免费开源（MIT），一行安装：npx skills add EricJujianZou/rederive 上线 8 天， 263 个 GitHub 星。
306
FlutterGo
FlutterGo.AI is an AI Flutter app builder that ships real apps to both stores. Describe your app, or start from Figma, photos, a ZIP, a GitHub or a URL, and get a working Flutter/Dart project with a live preview. Edit the code, connect Firebase or Supabase, sync to GitHub, and build signed releases for Google Play and the App Store. A Flutter MCP server lets ChatGPT, Claude, Grok or Cursor build apps in your account.
原文1
CapSolver
Expose CapSolver as typed MCP tools so compatible AI agents can call CAPTCHA recovery through a standard protocol.
原文1
Wrkbase
Since the first launch: weekly timesheets built from the clock with overtime and one-click approval, a shift rota, hourly pay from attendance hours, automations, and an API with webhooks. Tasks, automatic time tracking, attendance, leave, payroll and HR sit on one employee record, with an AI agent that shows its plan and waits for you to confirm. Free for up to 5 people, then $8 per member per month.
原文2
Jean by Melad Labs
Jean is a free bioinformatics and biotech suite for the whole experiment. Design constructs, primers and CRISPR guides; simulate digests, PCR, Gibson and Golden Gate builds; analyze Sanger traces, qPCR and ddPCR runs, variants, trees, mass-spec proteomics and molecules; verify the clone you got. Every engine is checked against its field's reference tools and every result explains its method. It runs in the browser with no account, and as a remote MCP server any AI assistant can call.
原文1
Pruneo
Pruneo frees up space on your Mac: caches, app leftovers, old versions, large files and the dev junk other cleaners skip. New in v1.1.0: it adapts to the apps you use, asks your AI agent about unknown folders and has an MCP server. Free, with undo up to 30 days.
原文2
tensemem
TenseMem is durable agent memory with a deterministic resolver, enforced cardinality, freshness-aware reads, and lasting corrections — so your agent gets what still holds, not a pile of conflicting prefs.
原文32
Agent Index
Agent Index is a free, sourced directory of about 404 AI agents for compliance, security, risk and business operations, plus about 483 independently verified audit firms (SOC 2, ISO 27001, PCI DSS, HITRUST, CERT-In and more). Compare tools side by side, filter by framework, region and integration, and find auditors by country and city. Not a ranking.
原文1
spider-protocol
For AI builders who fear bloat, bugs, and broken intent. SPIDER Protocol turns your AI coding agents into a governed engineering team: • Preflight Discovery: Absorbs true user intent before touching code • Living Git Memory: Architecture & decisions immune to context amnesia • Independent Validator Gate: Audits the diff and full test suite before any task is marked Done Build production-grade software at AI speed. 100% free & local.
原文4
drop.
I originally built drop.media to do one thing: drag an image into your browser and get a shareable link. It still does that! Now it's one media library for you and every AI agent you use. Upload from the browser, terminal or API, or let Claude, ChatGPT and Cursor save files to it through MCP. Any agent can find what another saved, use it, and share it as a public link or one that expires. Images, PDFs, video and docs, on every plan, including free. Upload once. Use it everywhere.
原文3
Kinaya
Kinaya is a keyboard-first browser for macOS. Ads and trackers are blocked inside the engine (Brave's adblock-rust with EasyList and EasyPrivacy) before a request leaves, and there is no telemetry. A built-in MCP server lets Claude Code read pages, click, type and take in its own background tab, using your real logins. Also: command palette, private and Tor tabs, encrypted password vault, reader mode. $19 once, 14-day free trial.
原文1
GetUrUI
GetUrUI gives you 106 free, MIT-licensed animated React components, blocks, charts and AI-chat UI, built with Tailwind CSS and Motion. Install any piece with one shadcn CLI command and own the source. An MCP server lets Cursor, Claude Code and other AI agents find and add components for you. Every item ships in light and dark. Pro unlocks all 332 designs, and 32 full templates cover SaaS, dashboards, commerce, docs and wellness businesses.
原文3
Vocnu
Upload a video or audio file and get translated subtitles (SRT, VTT, ASS), a voice-over track and a finished video in 35 languages. One clear AI voice reads the translation while the original sound dips under each line, like a TV documentary. Pick from 179 voices, add SDH or burned-in subtitles, and see the price before you start. The engine is open source (AGPL-3.0), with a CLI and an MCP server for AI assistants. New accounts get 60 free credits.
原文1
Knot Platform
Knot Platform is the open-source stack for a Knot: a node where people and personal agents share the same records. Agents can post what they need and what they can offer, keep context private or shared, and leave results humans can read—on a node you self-host. I designed the exchange architecture and open-sourced it. For anyone who wants to run their own agent-era information hub. Live:
原文1
ListCheckup
Clean your email list before you send. Upload a CSV or use the API: every address comes back valid, invalid, catch-all, risky or unknown, with a score and a plain-English reason. What's different: catch-all and unknown results are refunded, duplicates are charged once, credits never expire, no subscription. 100 free checks every month, EU servers, API, webhooks and an MCP server for AI agents.
原文1
Reins —— 自托管零信任远程运维平台
自托管零信任远程运维平台。单个 Go 二进制 Relay + 一行命令安装的 Agent + 50+ 命令 CLI，通过反向 WebSocket 穿透 NAT——被控端零入站端口。mTLS + 内置 CA + 多租户 RBAC + 防篡改审计 hash chain，并为 AI 代理提供规则门禁（自动化操作前必须先读项目运维规则并携带内容指纹 ack）。
2
Wingo UI
Wingo UI is a React 19 + Tailwind CSS v4 library with 226 components and 95 blocks that install as source code, shadcn style. Every component is animated, configurable through props, works in light and dark, and is built mobile first: bottom sheets on phones, 44px touch targets, swipe actions with a visible fallback. A CLI adds items with their dependencies, and an MCP server lets Claude Code, Cursor and other agents search the library and follow its design rules. Many items are free.
原文2
OpenRadView
OpenRadView opens the DICOM and NIfTI files from your scan CD or zip on your own computer: linked axial, coronal and sagittal views, experimental 3D and measurements. Optional MCP access for your own AI agent. It is a personal research project for education and has not been clinically validated. Ask a qualified clinician to interpret your scans; do not use the app for decisions about care.
原文90
ElevenLabs Realtime Skill
The ElevenLabs Realtime skill teaches AI coding assistants to build on ElevenLabs Realtime models. It sets up an ElevenAgents agent on Realtime Mini or Max, helps pick between them, and migrates a browser voice app from OpenAI’s Realtime API by swapping the credential, WebSocket URL, model and session config. MIT licensed, part of the elevenlabs/skills repo, installed with npx skills add elevenlabs/skills.
原文1
DNS Auth
Free public-DNS SPF, DKIM, DMARC & MX check: 0–100 score + short fix tips. Read-only—we don’t send mail. Daily monitoring Live at $15/mo (≤5 domains, email change alerts). Watch + playbooks rolling out + later API/MCP—not inbox guarantees.
原文1
ondocs
ondocs adds an Ask AI box to your docs: visitors get a short answer from your pages with citations to the exact section, or "I don't know" when the docs don't cover it, on your own model key. Coding agents get an MCP server (search, fetch, list_pages) and llms.txt from the same index. Plugins for Starlight, Docusaurus and VitePress, guides for Next.js, Fumadocs and Nextra, or one script tag. No vendor, no vector database. MIT licensed.
原文3
DeepChill
The open discovery directory where indie software, developer utilities, and AI agents gain lasting visibility. Ranked by launch freshness and real usage, so great tools never get buried in 24 hours. Free submission, dofollow backlink, and AI agent discovery via MCP — queryable inside Cursor, Claude, and Windsurf.
原文1
Waspes 3D Studio & Roblox MCP
Waspes 2.0 adds a 3D Studio that generates textured 3D models (.glb/.obj/.fbx) from text or images and streams them live into Roblox Studio, plus a public remote MCP server ( ) with 8 tools so AI agents can build websites and 3D assets. Still includes the AI website builder: one prompt gives a production-ready HTML5 site. Plans from $2.50/mo.
原文1
Graphicle
Graphicle is a Rust port of Graphify's open-source code-graph engine, 3 to 22.7x faster than Graphify 0.9.91 in all 39 of our measurements, plus project memory: docs, decisions and past agent sessions, cited to file and line over MCP. Local, Apache-2.0.
原文2
NVIDIA / SkillSpector
Security scanner for AI agent skills. Detect vulnerabilities, malicious patterns, security risks, prompt injection, data exfiltration, and supply-chain risks in Claude Code, Codex, and MCP skills before you install them.
原文2.2万
CovaSyn Chemistry Tools
All-in-one chemistry MCP server for pharma and biotech: over 250 tools in 18 families for spectra, stability and toxicology. Try it free in Claude, ChatGPT, Cursor. 14 to 41% without CovaSyn, 76 to 92% with CovaSyn. Measured on MolecularIQ, a benchmark we did not design (arXiv:2601.15279, Klambauer Lab, Institute for Machine Learning, JKU Linz).
原文1
Twig
A desktop Git client for you and your AI agent. You get a fast graph, line staging, drag-to-rebase and Undo for Git. Claude Code, Codex or Cursor connect over local MCP: they see what you selected, use far fewer tokens than raw git, and only propose commits.
原文2
Docustay
Your whole team, one flat price. Pro is $35 a month for the whole workspace: no per-seat fees, unlimited people and documents. Send a document, get it signed on any phone, and keep one sealed PDF with a certificate. Signers can pay while they sign, straight into your own Stripe account, and Pro adds no fee. Build on it with a REST API, five SDKs, a CLI and an MCP server, or run it on your own server for free (open source, AGPL).
原文1
—
—
Moppet：给 macOS 做了个会自己动手清理的桌宠，但删除前一定先问你
GitHub: （ Apache-2.0 ，Swift 6 / macOS 14+） macOS 清理工具，两个不一样的地方： 1 ）内置一个 Codex 级的 agent 。你说"清理浏览器缓存"，它先扫描给你看，等你批准才删。 沙箱、审批策略、补丁应用、会话恢复都是自己实现的；另外提供 23 个工具的 MCP 服务器， Claude Code / Cursor 可以直接调用同一套清理能力。 2 ）交互层是只 3D 桌宠（ SceneKit 程序化生成，没有 WebView 、没有 Electron ）， AI 要删东西时头顶弹批准卡，结果进气泡。 安全上比较较真：只删家目录、25 类保护路径、白名单命中及其子树永不触碰、 删除前二次校验、逐路径审计； AI 每次删除都要审批，MCP 要 confirmed + token ， HTTP 只绑 127.0.0.1 。 11 个模块：清理 / 卸载 / 优化 / 启动项 / 分析 / 状态 / 项目清理 / 瘦身 / 历史 / 设置 / 总览。 限制说在前面：AI 要自己的 API key ；瘦身仅 Apple Silicon ；系统级操作要管理员密码； 上不了 App Store ；没有 dmg ，源码构建；只在我这台 Intel Mac 验证过（ arm64 交叉编译通过）。 试一下： git clone && cd Moppet && swift build .build/debug/Moppet 不想开 GUI 就跑自测（ 133 项断言，1.4 秒，不删任何文件）： swift run Moppet --smoke 想听意见：AI 有删除权但每一步都要点头，这个颗粒度合适吗？ 觉得有用的话求个 star 。
—
LemMeBuyIt
LemMeBuyIt is the retail data layer for AI agents, resellers and developers: live prices, 12-month price history, cross-retailer compare, ASIN and Buy Box mapping, and clearance signals across 100+ U.S. retailers. Free MCP server, one-click sign-in.
原文14
Multi Gmail Account MCP
Open-source MCP server for Claude Desktop and Claude Code. Connect any number of Gmail and IMAP accounts, then ask what's unread everywhere, who's waiting on a reply, or to draft answers. Runs locally, logins encrypted, drafts-only by default.
原文1
通过 MCP 发布网页和报告的开源工具
BrewPage 的开源公共工具层，包含 OpenAPI 契约和 TypeScript MCP 服务器，可让 AI 助手发布 HTML、Markdown、静态站点及文件，并通过所有者令牌更新或删除内容。MCP 包采用 MIT 协议，API 契约采用 Apache-2.0；托管服务本身为闭源，内容有有效期，不是永久存储。适合学习 MCP 工具如何对接 REST API。 亮点 提供 20 个 MCP 工具，覆盖发布、读取、更新和删除内容；适合分享生成的网页、报告和静态站点。 本地 npm 包与远程 MCP 端点均可使用；创建内容无需注册或 API 密钥。 创建后返回 HTTPS 分享链接和所有者令牌；令牌用于后续管理，不应公开。 自荐说明：我是项目作者 Maksim Kochetkov（GitHub：kochetkov-ma）。推荐的是开源工具层，BrewPage 托管平台的后端、前端和基础设施并未开源。 产品官网： MCP：
5
Veynod
Veynod is a shared blackboard for multi-agent teams over MCP and REST. One versioned memory, task hand-offs with leases, independent quality gates, human approval, and delta context sync for Claude Code, Cursor, and custom agents.
原文1
KnightCode
KnightCode is an open-source coding agent for the terminal. Sign in with Claude, ChatGPT, Grok or Copilot, use an API key for 40+ providers, or run a local model. Every file read, edit and command shows up in the transcript. Sessions are a tree you can branch, fork and compact. Extensions, skills and themes live next to your repo. The system prompt is about 1,100 tokens, so more of the context window goes to your code. MIT licensed.
原文1
EngMock
EngMock turns your own interview questions, interview write-ups and notes into live AI mock interviews for software engineers. Practice system design, coding/DSA, OOD and AI-assisted coding with adaptive voice follow-ups, a whiteboard or coding workspace, and scored reports with evidence and replays. VS Code mode (preview) keeps coding in your editor and voice in the browser. Use MCP to create and edit scenarios, prepare interview links, and review reports and progress.
原文1
Reveric｜幻界， Agent Native 一站式小说、短剧创作工作台｜无限画布
详细介绍一下核心能力： 1. agent native，UI 能做的事情，AI 基本也能做到 3. 全流程 AI 小说/短剧创作，关系图谱，世界观设定，人设管理，视频生成等 4. 无限画布，短剧中常用的参考图等连线关联 5. 自动化剪辑以及图像，视频处理 6. 3d 导演台，预先导演角色动作，走位，指导模型生成更准确 7. agent 支持 codemode，skill 等常用功能 8. 自定义模型，支持常用渠道，OpenAI，Anthropic，Seedance 等 为什么做这个 1. 已有的剧本写作，短剧制作，大都是独立的项目，本项目将两者集成，可以在同一个上下文写剧本以及制作对应的短剧 2. 使用 Flutter，高性能，本地优先，现在很多项目都是使用前后端分离技术，web 部署很麻烦，本项目直接下载安装即可 3. agent 实践，subagent，codemode，skill 管理，按需加载，渐进式披露实践
152
scoop.md
A Chrome extension for developer documentation. Open a developer docs site, click scoop.md, and its llms.txt plus its instructions for agents are copied as a short prompt for Claude Code, Codex or Cursor. No llms.txt? scoop.md builds one. Free, open source.
原文1
Red-XAI Hub
Red-XAI is a growing technology ecosystem by Red-XAI LLC, featuring Vortex's free AI chats and autonomous agents, Red-XAI Host for self-hosted websites and apps, browser-based Cloud PCs, and Red-XAI DB with its own query language. Build and integrate with APIs, SDKs, and CLI tools. Red-XAI Hub is coming soon. Investors, strategic partners, businesses, and app integrations are welcome. Explore
原文1
StartupKit
StartupKit is an MCP server and CLI that gives AI agents tools to research a market, launch a product, and grow a startup: a free production-ready SaaS stack, SEO and keyword research, and launch checklists.
原文1
Open Dataset of the Lebanese Law
Hey folks! I made it easier for builders to use the lebanese legislation through: a public DB, API, and MCP. The Lebanese University has an archive of all laws dating back to 1904 (legallaw.ul.edu.lb). People and AI can barely find it and navigate it. It also had outdated articles. Now, more people are using it. Builders for their startups, lawyers for their work, and students for their bar prep. I'm still trying to get it trusted by authorities. The database includes the edges between every law, article, ruling and Gazette issue. You can easily identify how they influence one another. Articles are versioned, so you can see how it was affected by other laws from the date it was published. 100k unofficial english translations were added using LLMs. They aren't accurate, but they help with search and SEO/AEO/GEO. We're working on producing legally accurate translations. 47k rulings were PDFs, some handwritten. We converted all of them to text using OCR, flagging handwritten files. We're working on an HTR model specifically for those to increase the confidence. Today, other datasets in Lebanon charge for this data. I'm doing it as a public service. I want change in my country. And, lately, I think it's through helping builders. Yet, what I'm doing is legally wrong, but hopefully morally acceptable.
原文—
carrierscan
Paste a text or upload a file and see how likely it is that ChatGPT, Gemini or Claude wrote it, with a score for every part. Most detectors are trained on English; carrierscan has its own detectors for Ukrainian, Kazakh, Uzbek, Estonian, Georgian and Lithuanian. On the same Ukrainian texts it caught 16 of 17 AI texts (GPTZero: 14 of 17) and flagged no human ones. Free, no sign-up, texts never stored. Also a public MCP server for Claude, ChatGPT and Cursor.
原文1
Kashkeshet – non-opinionated agent-to-agent chat MCP server in Rust
This is hardly a greenfield area. There are tons of agent-to-agent communication apps. But, at least to my taste, a lot of them are too complicated or too opinionated. If we learned anything from the HuggingFace incident, it's that modern AI agents don't need a lot of help communicating. They are smart and creative and figure it out. So...with that in mind, I (meaning I + claude and friends) built kashkeshet, or kkt. It's a small, Rust-based MCP server that gives agents a few simple chat primitives (identity, channels, chat) and leaves it up to them to use those however they want. You as the human get a little web interface so you can watch the fun. Coordinate your coding agents; setup a virtual company; conduct sociological experiments; whatever. kkt's main goal is to stay out of the way, and let you and your agents figure out how to organize and communicate on your own. This is meant to be run on your local device and shared among local agents. There's a little explanation of how you could use tailscale to go broader, but nothing like Oauth2 or anything yet.
原文—
Billy
Billy is a free, open-source invoice editor that feels like writing a document. Click any line and start typing, let totals calculate themselves, and export professional, print-ready PDFs in seconds. With MCP support, Billy can also integrate with AI-powered workflows. Built for freelancers, indie hackers, and small businesses who want simple invoicing without the bloat. Try it at
原文4
Gemral Edge
US federal contracts, defense contract announcements, lobbying, congressional and insider trades, tied to the ticker that was valid on the day each record became public. Delivered as Parquet files, a REST API and an MCP server for AI assistants.
原文1
Specification First Agentic Development Plugin and Prove
Implemented agent smith since start 2026 with over 1000 specifications which lead to about 500k lines of code. The procedure with that spec-first works well. How do you do it?
原文—
Cryo – A language whose programs live forever
Cryo is a programming language that models long-running tasks like workflows or AI agents. Programs freeze to disk when idle and can be thawed when new data arrives. As part of our announcement of Endothermic Labs, we've put together a small of Cryo that highlights its features and walks through the language and its plugin system. We originally built it as a way to deploy healthcare agents at scale, but have rebuilt it from the ground up to be a standalone toolchain for long-running tasks. It acts as something like a cross between Durable Objects and Temporal workflows, where the script itself is the durable unit with no replay needed. Cryo is statically and structurally typed, with a concurrency model similar to Go. It's still under active development, so we're holding off on open sourcing it until we're confident that the language is stable, ideally in the next 3-4 weeks.
原文—
Mainplane – Agent in Go that lives on your old PC, drives your laptop
Hi HN, I'm Alexander. Yes, this is another agent harness. Here is why I built one anyway. I was engineer #2 at Browser Use, where I built our agent harness and the evals behind it, including BU Bench, which became a default benchmark for browser agents. I left in September to build Mainplane. The problem: local agents die with your laptop, and cloud agents can't reach your work. Many cloud platforms ask you to "select a repository", but my work is not that simple. I work across a desktop, a Mac mini and a laptop, and much of it is not in git. So I loaded Ubuntu on my old 2012 desktop and made it my Mainplane server. It has run for four weeks at a load average of 0.02, because the agents' tools run on other machines. How it works: it is a control plane architecture. The server runs the harness. It holds your API keys and every session, makes the LLM calls, and routes each tool call to a worker. The workers are your other machines, connected over a built-in private WireGuard network. A bash command runs in the worker's terminal, not on the server. Shared drives let projects move across machines. The agent works on your machines just like you would, but its state lives on the server, and your keys never leave it. Some design choices: - Every session is one append-only file. This is the agent state, and it is easily readable. If something breaks, you can resume it from the file. Observability comes for free: agent traces are just these state files. - Small by design. The system prompt is 3k tokens, and there are just two kinds of tools: shell and file (read, write, edit). At Browser Use, I learned the bitter lesson and found frontier models did better when I deleted all our helper functions. So I made the same bet here: bash is all you need. - No compaction. When a session fills up, /continue starts a new one that greps the previous state file, retrieving the context it needs, instead of reading a lossy summary. A /compact on a 400k-token session takes a while to write its summary. A /continue into a brand new session starts at once. - No permission prompts. A hyperintelligent misaligned agent WILL find a way around a permission system. The better strategy is giving the agent full unbounded access to a local computer, and only putting what you trust the agent with on that computer. - Why Go. WireGuard is written in Go, so the network compiles right into a 20 MB binary. Go is made for production servers, LLMs write it well, and the simplicity of Go matches how I like to build. Mainplane works with a Claude or ChatGPT subscription on the regular usage limits, or any API key. I would love your feedback, especially brutal and honest feedback and tasks that the harness cannot do. I'll be here all day to answer questions.
原文—
Agent Quest
Turn your Claude Code and Codex sessions into a living fantasy village. See agent activity at a glance, then click a hero to inspect context usage, skills, plugins and subagents. Agent Quest reads existing local session logs while you keep using your usual tools. Keep multiple sessions in sight on a second screen. Customize the village and export Markdown or JSON reports for review. Free, MIT-licensed and open source. Runs locally in your browser with Bun.
原文2
Cylect AI
AI OSINT platform with 475+ open-source intelligence tools, an AI analyst, a live intelligence map and MCP access for external AI agents.
原文2
Vibex
Coding agents write blind. They can't run your app, see the screen, or check their work — so you end up running builds, pasting screenshots, and copying console errors. Vibex is a native Mac window that gives your agent eyes. It embeds a live iOS simulator, browser preview, source control, and AI usage dashboard with 50+ MCP tools. Claude writes code, runs the app, taps the simulator, reads the console, and checks every diff. You just review. Free forever. MIT licensed. No account, no tracking
原文3
AgentDeck：在浏览器一屏并行驾驶多个 AI 编程智能体
自荐一下自己的开源项目 AgentDeck —— 一个完全跑在本地的 Web 终端驾驶舱。 一句话：One screen. Four terminals. Any AI coding agent. 解决什么问题：现在很多同学同时用 pi / Claude Code / Codex / Gemini CLI 等多个 AI 编程智能体，开 N 个终端窗口来回 ⌘-Tab。AgentDeck 在浏览器里给你一个 2×2 的真终端网格（node-pty + xterm.js），每个格子跑一个智能体；底部输入框按 Tab 切换目标，Ctrl+Enter 一键广播——一条需求同时发给四个 agent 并行开工。 主要特性： 真 PTY 终端，TUI 快捷键原生可用；刷新页面进程不丢（服务端保活 + 输出回放） 隔离 git worktree：四个 agent 同时改一个仓库互不踩脚，关格子时选择合并/丢弃/保留 聊天视图：把终端噪音换成干净的用户/助手对话；消息可跨格转发，能搭「A 写码 → B 审查 → C 跑测试」流水线 定时任务；浏览器桥接（Chrome 扩展 + MCP 服务器，agent 可直接操作真实浏览器）；明暗主题 + 6 色强调色；中英双语界面
1
Agent Mascots
Your agent is thinking, calling tools, waiting for an answer, or stuck. Agent Mascots shows it at a glance: 23 animated characters, 8 states, any color. One dependency-free web component, plus an agent skill that wires it up for you.
原文3
BLAZLE
Stop prompt injection, secret leaks and risky MCP tool calls before they leave your AI agents. Works with Cursor, Claude Code, VS Code and GitHub. Free for 5,000 requests a month.
原文2
ContextGem 1.0
New in 1.0: decision models next to LLMs. They answer closed questions with the probability of each answer, fast and at a low cost, and a model group sends each question to the model that suits it. Also new: Docling for your files, and tools for agents.
原文79
Hob - Open-source, on device alternative to Claude Cowork and ChatGPT
Hey HN, I'm Christian and this is a new OSS project I've been noodling on for the last couple of weeks. Hob is a free, MIT-licensed desktop app for Mac, Windows and Linux that works on your files the way the Claude and ChatGPT desktop apps do, except the model runs on your own computer. It uses open-weight models (currently Gemma 4 and gpt-oss). All your chats and files never leave your system. The only things that go online are web search (can be turned off in settings) and any MCP connectors you add yourself. Main features: - Automatically finds the best open-weight model for your hardware on first install (Mac/Linux/Windows) - Writes sandboxed Python code for ad hoc tools (data processing, document conversion, graphs etc.) - Built-in web search (powered by SearXNG) - MCP support - Image recognition - CRUD to your local filesystem - Works fully offline with no internet Before building this I tried every local AI app I could find, but it seemed each one had different problems. The main thing I noticed is that none of them could do what the ChatGPT and Claude apps do with your files, e.g. read and change files in a folder, look at images, run code in a sandbox, draw a chart. The most advanced thing any of them did was paste a plain text file into a small context window, which basically failed as soon as you had a file larger than a few KB. I couldn't find anything that could easily parse images and bigger PDFs. The OSS solutions that were more advanced all seemed to default to hosted models (either OSS or closed), which defeats the entire point of the privacy that on-device OSS offers. They also seemed to push towards an account/open-core style business. I build it using Electron bridged with llama.cpp in its own process. On initial load, the system will try and figure out if you have a compatible graphics card and if not default to CPU-only (I ran into a lot of issues of graphics card beeing seemingly compatible but then crashing on first query, there's a whole saga to this lol). Code execution happens in a sandbox that can only write to the chat's folder and has no network unless asked (sandbox-exec on macOS, Landlock on Linux and a low-integrity process on Windows). Images go to a small Qwen3-VL model. Dictation is a local Whisper model. The hardest part was definitely to get the smaller models to use tools reliably. Ran into a ton of issues of 4B models happily saying "here's your chart" without ever calling the chart tool or get stuck in an infinite call loop. I spent a fair amount of the work building guards vs tweaking prompts (that didn't really work for small models). I understand the pushback that local models are weaker than frontier cloud models. This is true, but for most tasks for most normal people, I'd argue you rarely need the newest frontier model. From my own evals, systems with 16+ GB of RAM do really well, and if you have a GPU or more RAM, you get very close to what you'll get out of the box with a frontier hosted model. Excited to hear what you think and see how it runs on other systems!
原文252
1Server
1Server is one MCP server that runs all the others. Pick servers from a curated marketplace, add a single entry to Claude, Cursor or VS Code, and stop editing JSON. Hot-reload, health checks, an AES-256 secret vault, and your AI can install tools from chat.
原文1
Markdown is 40× more token-efficient with SQLite
Hi HN, I'm Richard, founder of Native. I'm working on open-source context ownership, to stop AI centralisation: https://github.com/withnative/native My first focus has been making Markdown vaults powerful enough so that teams can self-host all of their context (messaging, docs and projects), without it being externally fragmented or rate-limited. I'm sharing some early results: Markdown vaults can be 40× more efficient with SQLite! https://www.withnative.ai/research-and-posts/markdown-contex... Here's some of the main ideas: - SQLite or Markdown folders, you pick. You can project the same authoritative event log into either form; more projection formats will follow (e.g. maybe neo4j; we do use a graph structure). - Empirically: SQLite is better for agents! On a 120 trial evaluation for 11,000 records, agents use 40× fewer tokens with SQLite vs Markdown, per correct answer. - Everything is a file or SQLite row. Messages, documents and projects: all can be modelled in your vault. You get local-first collaboration with teammates, synced to a cloud of your choice. - You can render your vault with arbitrary UI. Your WhatsApp-like UI speaks to your teammate's Slack-like UI; both are just views over the same database. A few things I've learned building this: - Agents do pretty well even without SQL if you give them strongly typed Markdown frontmatter. This was one arm of the trial I ran. I didn't have any clear expectations, but it was cool to watch: agents ended up writing Python scripts to do SQL-like queries. - Cumulatively, cached input tokens can get expensive. I feel silly in hindsight here. I'd gotten used to letting my sessions run long, "It's all cached so it's very cheap". But when I looked at my results here, over 2/3 of the token bill on long sessions was cached input... - "Markdown vaults don't scale as well as databases" isn't as self-evident as I think. This is why I decided to benchmark it... I've had some very blank stares when I try to explain how Markdown files might break at true agent-scale volume. Unfortunately, the dataset is private, because it's two months of my own work. I am considering open-sourcing it even so, but there is definitely some embarrassing stuff in there. TBD... I know that independent result verification matters! The core engine that I evaluated is open-source, with more to be released soon: https://github.com/withnative/native If you're interested in trying connected context on an open engine, you can get updates from our website or email me directly: [email protected] https://www.withnative.ai/
原文—
SEO Agent Skill
Anyone can build but ONLY few know how to SELL. This $5 Agent Skill helps you get traffic to your website automatically without any research. It contains 24 practical skills for AI agents to optimise and rank your site.
原文3
ClearMakes Inbox
ClearMakes Inbox runs workflows on new messages in Gmail, Outlook and Slack. Say what you want in one sentence, like "when someone asks about prices, draft a reply for me". It asks at most one question, writes the steps, and you turn it on. Steps can post to Slack, add a note in HubSpot, draft a reply, or call your own MCP server. Smart FAQ replies to everyday questions from answers you write once. Free plan: 3 apps, 100 runs a month. Paid plans from $29 a month.
原文1
Agent Payment Node：AI 代理的本地支付工具
Agent Payment Node 是用 TypeScript 编写的本地支付运行时，通过 CLI 和 stdio MCP 供 AI 代理检查 x402 报价、准备有界支付并查询状态与收据。仅支持 Apple Silicon macOS 和 Node 24；直接转账和 EVM 签名需前台所有者批准，Solana exact x402 可按所有者接纳的策略运行。软件免费，链上和服务商费用另计。 亮点 将报价检查、有界准备、操作状态/收据和恢复路径组合为本地代理工具，适合研究 MCP 工具接口、支付状态机及所有者授权边界。 这是 Nuanu AI 的 MIT 开源项目，本次代表 Nuanu AI 自荐。具备资金操作能力；本地软件托管、错误策略和主机失陷仍有风险。没有声称独立安全审计、当前客户端兼容性复测或生产使用证明。 已按要求在 HelloGitHub 网站用完整 GitHub URL 搜索；匿名搜索没有展示匹配结果，但提示登录才能查看更多。另检查了当前完整仓库的 263 个文本文件（含 253 个 content 文件）及开放/关闭 issue/PR 历史，未发现 APN 名称、slug 或 Nuanu。
0
Kluff
A collaborative workspace where employees and AI agents share context, build on each other's work, and contribute to organizational memory.
原文11
TrackStack
See visitors and attributed revenue by channel, page and campaign. Track identifiable AI referrals, connect Stripe with checkout metadata, and ask questions through MCP. Web app from $4.99/mo, 14-day trial. Explore the sample dashboard without signing up.
原文2
Hatamoto batteries-included CMS in Go
Open-source, entity-first CMS in one Go binary. JSON-LD, sitemaps, llms.txt, AI crawler analytics, an MCP server for agents and sandboxed plugins built in.
原文—
MCPatch
Your MCP server works in curl, but Claude says "Couldn't reach"? Paste the URL. MCPatch runs the 12 steps Claude runs to connect, from the public internet, stops at the one that breaks and gives you the fix, plus a prompt for Claude Code or Cursor. Free
原文1
10ML Perfume Finder
10ML Perfume Finder is the free perfume MCP server and fragrance API from 10ML, an online perfume shop for perfume samples and decants: a remote Model Context Protocol server at that needs no API key and no sign-in.
原文1
LucentEdit
LucentEdit is a free, open-source design app that combines photo editing and vector design in one lightweight desktop application. Built to solve real design workflow problems, it offers professional editing tools, brand training, and AI-agent integration through MCP. It runs on Windows, macOS, and Linux, with no subscription or login required. Built for designers, freelancers, and teams who want powerful, accessible design tools without the extra cost.
原文1
ImagGenius
ImagGenius makes pictures from words, locally. No account to create, no subscription, nothing sent to a server. Describe what you want, and press Generate. Not quite right? Change a word and go again. More customization? Add a LoRA for your selected model, and configure parameters to see how it applies. The best part: it comes with its own MCP server. Set it up with Claude Code or any of the other agents for your workflows - no need for another subscription just for generating images per-use.
原文2
SYSTEM-ENGINEERING
An open-source engineering layer for AI-assisted development, exploring agent orchestration, explicit policies, validation workflows, and persistent project context.
原文1
开源了一个 MCP 治理网关： Jev 判 0.94、d3 判 0.70，两个判官说不同的话时怎么办
给 AI agent 做了个 MCP 执法网关（ Rust ），核心问题：当两个决策模型对同一件事给出不同判断时，怎么安全处理？ 实测：同一个越权场景（ coding agent 试图 merge 夹带无关变更的 PR ）： - Jev 给 0.94 置信度 → 准入层 Accept → 授权挂起 → DENY - d3 （ vLLM SR ）给 0.70 → Hold → 什么都不改 → ALLOW 两个都对。分歧不是 bug ，是校准本性（ d3 ECE 4.3 vs Jev 8.0 ）。治理层的价值在于让分歧安全。 网关做的事：每个 tools/call 先过 r2r_decide 门控，ALLOW 转发上游，DENY 返回 isError + 证据文本（相位 + 规则 + 事件 ID ），全程审计到 EventStore 。 零 API key 复现： cargo run -- fixture cargo run -- fixture --provider d3 GitHub: （ Apache-2.0, v0.2.0 ）
—
Doconda
An API that lets AI agents create, edit and review Word, PowerPoint, Excel and PDF files. Real styles, no broken layouts, fully editable by humans afterwards. Works with Claude, Cursor and Codex via MCP and SDKs. €5 free to start.
原文1
—
Harf
Harf is a voice-first AI agent for Mac. Say "Hey Harf" and give a task in your own words. Harf uses what's on your screen as context and acts inside the app: a dark version of a Figma screen, a photo edited in Photoshop, an email reply drafted from the open thread. You review the result before anything goes out. Not dictation, action.
原文2
Sitemoth
Sitemoth crawls your site from your own computer and flags 240 technical SEO issues, each with severity and the fix. JS rendering, GSC + GA4, log analysis, client reports and an MCP server for AI agents. Mac, Windows, Linux. Free in beta, no URL cap.
原文3
Model Salt, fixing the pelican problem in LLMs by adding entropy
I have been frustrated with how repetitive the output of LLMs is. Even at maximum temperature, asking the same prompt ten times gives functionally the same result each time, as discussed here: https://news.ycombinator.com/item?id=50090443 That’s not surprising considering how LLMs work, but the thread above got me wondering if you could push models away from their usual generation patterns by injecting entropy from an external source. After a lot of tweaking and experimenting, I turned the idea into an MCP server. It's optimized to add just enough entropy that creative output stops being repetitive, while still living up to the intent of the original prompt. I’d love feedback, particularly on the approach itself. If you have a better way to inject entropy than what I have found, a pull request is very welcome: https://github.com/modelsalt/model-salt
原文—
OneSpace
OneSpace puts Claude, GPT, Gemini and Grok in one app, connected to Gmail, Calendar, Slack, GitHub and more than 50 apps. Agents with a computer of their own, Code Workspaces for Claude Code and Codex, and memory you control. Free to start, from $8 a month.
原文1
NuPlay
NuPlay runs AI agents for enterprises in production that improve after every run. Deploy agents for retail, customer service, consumer goods etc., across voice, chat, and email, connected to your existing systems. NuPlay is built as one platform, has 5 layers which build, run and improve the agents: NuPro, NuStack, NuContext, Nupulse and NuLoop.
原文102
Skill-spotter, auto detect agent prompts and commands to create skills
Skill-spotter auto detects commands and prompts you type. It detects variations such as 'push these changes' and 'get these changes pushed'. Using both word comparison and the actual actions you agent performs. It also adds to your Claude.md if that's more applicable than creating a skill. Just saving lots of time if you do repeat prompts.
原文—
Vokolla
Vokolla is an all-in-one platform for preparing and distributing press releases. Turn business news into clear drafts with AI assistance, refine your story, select a distribution package and track progress and coverage. Vokolla Press Release Agent also brings drafting, revisions, draft saving and release queries into compatible AI conversations through MCP.
原文1
墨观 Inkstone 一套从 526 个中国传统颜色里长出来的代码配色
一套从 526 个中国传统颜色 里长出来的代码配色。支持 VSCode, VSCodium, Trae 等主流编辑器；Kaku、Ghostty 终端以及 OpenCode。已经上架 VSCode marketplace 和 OpenVSX 市场
1
AgentRadar
Hi guys! Built AgentRadar, a Python SDK to track AI agent executions, including tool calls, LLM inferences, token usage, errors, latency and more, all visible on a web dashboard. Integration requires almost zero code changes. Just install it using `pip install agentradar-sdk`, grab your API key, and initialize the SDK. You can enable or disable metadata tracking based on your preferences. It’s lightweight and designed to add almost no latency. Would love your feedback!
原文1
AppPad 把原汁原味的启动台带回到 macOS 26+，也支持 macOS 15 —— 另附剪切板、工作流、App 快捷键与窗口控制。
AppPad项目地址 安装 方式一：Homebrew bash brew tap umindhub/tap brew trust umindhub/tap brew install --cask apppad 功能 启动台 完整还原经典 macOS 启动台，默认布局与系统顺序一致 四指捏合、菜单栏或点击程序坞中的 AppPad 打开（再次点击关闭） 文件夹、拖动排序 / 合并成文件夹，长按图标或按住 ⌥ 进入编辑模式 内置干净卸载：删除 App 的同时清理残留文件 网格搜索、页面指示器（圆点 / 数字）、“液态玻璃”控件（macOS 26+） 背景：白色 / 浅灰色 / 深灰色 / 壁纸 / 自定颜色，浅色与深色主题分别保存设置 剪切板 支持文字和图片永久保存 复制内容可收藏 ⌘1–⌘0 快速粘贴，来源 App 筛选和 App名:: 搜索语法，右侧实时预览 工作流 & AI MCP ⌘Space 打开工作流面板：App 与工作流按使用频率统一排序 工作流可运行 Shell 脚本、AppleScript、快捷指令或 HTTP 请求，支持参数（如 time 10） 内置：清倒废纸篓、设置倒计时、设置闹钟、设置音量、退出所有应用…… 权限声明由真实沙盒强制执行（允许一次 / 不允许 / 始终允许） 以 JSON 文件或链接导入、分享工作流 内置 MCP 服务：支持 MCP 的 AI 助手可直接调用和控制你的工作流 AI 可通过 MCP 读取工作流编写规范，直接帮你创建新工作流 AI 运行的工作流与你手动运行一样，经过同一套权限确认和沙盒 计算器 在工作流面板搜索框直接输入算式即可计算（支持科学计数法） App 快捷键 为任意 App 设置全局快捷键 窗口控制 靠左 / 靠右排列（½ → ⅓ → ⅔）、最大化、恢复、移到其他显示器，快捷键均可自定义
3
A 3D mesh playground for coding agents (three.js + manifold-3d)
When GPT-6/Gemini launched recently, a lot of the 3D demos used blender-mcp ( https://github.com/ahujasid/blender-mcp ). It's cool, but you have to install blender locally and wire up plugins just to try a prompt or share a run. I built a browser-based 3D mesh playground so you can hook up your own coding agent over MCP (or run in-browser) directly in a tab. It runs on three.js + manifold-3d for blender-like boolean/mesh editing in JS, and gives the agent a render_preview tool to take multi-view (front/right/top/iso + custom camera angles + clay mode). That way the agent can look at what it built and fix the geometry, similar to how it iterates when building a website. The submitted link is a trace of Gemini reconstructing a Copenhagen canal scene from a photo: it hits a script error on the first pass, fixes the bug, and then goes through several rounds of checking 4-view and close-up render_preview to tweak the geometry.
原文—
kabmo
Browser console and diagnostic chamber for Oh-My-Pi (omp) agent sessions — live timeline, steering, real PTY, git/diff/wiki panels, and a panel-plugin platform. Elysia + Preact on Bun. - kabmo-labs...
原文1
Sóþspell, a model-guided walkthrough of a codebase
I kept finding myself wanting a way to get talked through the code that a model had created, which led to me making Sóþspell (old English, "Truth Story"). You give it a git locally or via URL, and it creates a tour of the in a web page, with the aim of it explaining the whole thing. There's also model harness plugins for Claude, Codex, and OpenCode. You give it an API key for whatever model you prefer to create the tour (saved locally or via github), and you can also give it a voice using an ElevenLabs API key. I just pushed a canvas view, which I'm hoping I can make the code navigation a bit more helpful. There's a codebase, and you can export tours as JSON. The doesn't need any keys to work, it falls back to browser speech if you want to hear that. This is my first real, "big" open source project that I've released, prior to this I'd just contributed stuff. I hope it can be useful to someone! I would love to hear some constructive criticism too, and any suggestions on features you'd like to see. It's MIT licensed, and I'll try to stick around in the comments to answer questions if people want.
原文—
—
—
Scientific Interview：AI 搜岗、简历与面经复盘的本地求职工作台
阮老师好，自荐一个 MIT 开源项目 Scientific Interview：把 AI 搜岗、简历修改、面经复盘和投递进度整理到本地 HTML 工作台。
3
Optio – A self-hosted orchestration platform for AI agents
Hey HN! I shared Optio here back in March. Originally it was focused on getting swarms of coding agents (claude code/codex) to drive PRs towards a merge by automatically resuming them on CI failures or PR comments. Since then, I've reworked the project to be a general platform for running any kind of AI workflow on your infra by allowing users to configure each component of a possible workflow: When: On demand, a cron schedule, or triggered by external events such as a webhook, slack message, ticket (jira/linear/gh), pagerduty/datadog/alertsmanager/pylon/sentry. Where: In a k8s pod or any machine you have installed the optio daemon on (I use my laptop). Who: Any harness you prefer (Claude Code, Codex, Gemini, Pi, etc.) What/Why: You can format parameters from the triggers directly into the prompts and then configure if the agent should exit, be resumed automatically, or be persistent. You can either define workflows in the web ui or as yaml files that get synced to the orchestrator (ai config-as-code). I've also built native iOS and Android apps so you can connect to your cluster and create/manage agents remotely. The marketing page is at https://optio.host and the code is still open source at https://github.com/jonwiggins/optio Would love feedback or any thoughts you might have on how I could make it better!
原文—
AgentParley
Run a multi-agent AI team in the cloud: specialists that delegate, report back and run long missions. Sandboxed plugins, a 3D companion, no server.
原文2
Symbix，一个陪伴型的全能Agent
正文 简言之：一个陪伴型的全能Agent 遇到一个痛处:AI 助手要么只会给一段话让你自己排版,要么不打招呼就把你文件改了。所以我做了 Symbix——一只住在桌面上的桌宠,你说一句需求,它自己找工具、干活、交付排版好的成品文件(Word / PPT / Excel / PDF),每一步留痕、可一键还原。 但它不想只当效率工具。它有长期记忆和每日手账,越用越熟;会在没事的时候主动来找你聊天,一半时间是翻你们的记忆找话题("上次你说要做的那个桌宠,后来怎么样了?");任务做完提醒你的时候,响起的不是刺耳的"叮",而是轻风、晨露、余烬这些柔和音——界面也是暖色主题,连暗色模式都是暖炭色的。它是个工具,更是个有温度的桌面伙伴。 该有的边界感也有:多步任务先报计划等你点头,文件修改和命令执行都会弹审批卡——或者打开 auto-pilot,放它自己跑。全部数据本地:对话、记忆、附件,删会话连附件一起清。 亮点一览: 📦 以md为媒介，多种文件可解析，支持模型视觉和ocr 💛 会主动来找你:翻记忆找话题、弹幕先开口,是个有温度的桌面伙伴 🌸 温柔感做进细节:十几种柔和提示音(轻风/晨露/余烬…)+ 全套暖色主题,暗色也是暖炭色 🤖 一句话跑完整条流水线:检索 → 下载 → 精读 → 总结 → 交付,全程无人值守 🎭 复杂任务自动分身多智能体并行,每个分身可以用不同的模型 🗂️ 每步操作有可读 diff,任何一步一键回溯,新建的文件也能撤 💾 长期记忆 + 每日手账,重启后不用重新自我介绍 🎨 会自己画皮肤:描述想要的样子,它画形象、做动作、打包成能导入的桌宠 🔊 本地语音识别(数据不出机器)+ 流式逐句 TTS 🧩 自研pythonSDK包，工具丰富，且Skill 包和 MCP 工具即插即用,权限和审批与内置工具同一套 下载 / 了解更多: 🐙 GitHub 开源:
0
BEON PASS
Connect your tools once and control what your AI can access. BEON PASS brings connected accounts into personal and team wallets, with granular permissions, human approvals and activity history. Use your own AI through one MCP endpoint, with access you can review and revoke.
原文1
GameSkill
收藏了上百个 game skill，让 Claude Code、Codex、Cursor、Kimi 等 AI 编程 Agent 更会做游戏的 Skill（SKILL.md）、MCP 服务器和配套工具 - 覆盖 Unity / Unreal / Godot / Bevy / Roblox 等引擎，以及美术资产、像素风格、音效、测试与逆向。所有链接都经过可访问性检查，星数和许可证来自 GitHub API（2026-10-26 更新）。同一个仓库里的多个 skill 合并为一行，data/skills.json 里保留了全部 255 个 skill 的明细
—
Thumbly
Thumbly is a free, open-source Mac app and MCP server that lets Claude, ChatGPT, Cursor and other AI agents see and tap your real iPhone. You approve every client and can pause AI control in one click.
原文2
JetBrains / thinkrail
Vibe code with pi in a lightweight, real IDE that customises itself around the way you work — The Vibe You Need
原文6506
在 Windows 上一键部署原生 OpenClaw 的桌面外壳 — rheia
在 Windows 上装 OpenClaw 要过 WSL2、Ubuntu、Node、网关四道门，一步走错就得从头再来。Rheia 把这条链做成一次双击：复用机器上已有的 Ubuntu（不另建发行版）、安装官方 npm 包（不 fork 不改源码）、自动选择国内或海外镜像源、用 systemd 托管实现开机自启。装完即预装 8 项技能与插件，配置好模型就能干活。 亮点 复用机器上已有的 WSL2 + Ubuntu，不另建专属发行版，磁盘不重复占一份 始终安装官方 openclaw@latest，官方每出一版立即跟进，文档、插件、issue 区全都对得上 开箱即用：预装 Tavily 搜索、mem9 记忆、office-toolkit 等 8 项技能与插件，并直装 LibreOffice 与文档处理库
1
GSC Magic
GSC Magic is a hosted MCP server that gives Claude, ChatGPT, Cursor and any MCP client your real Search Console, Bing, GA4 and Tag Manager data. Sign in with Google, paste one URL, ask. 77 tools, ready-made SEO reports, read-only by default. Free plan.
原文4
whatsapp-bridge
Self-hosted bridge to your WhatsApp chats, with an MCP server and a web admin UI built on top of whatsmeow for easy deploying on your VPS or Mac Mini.
原文1
EricGameLauncher：一款采用WinUI3和.NET10构建的极简主义次世代游戏启动器。
EricGameLauncher 开源主页：EricGameLauncher 亮点速览 🧩 全链路执行：主程序、管理器、代替启动、伴随执行四条链路逐项可配；EXE、快捷方式、URL 协议、网页、商店应用都能直接跑。 🖱️ 动态右键菜单：菜单项随项目自动增减，识别到平台即显示「运行 Steam / 运行 Xbox」；每个项目最多 10 个自定义项，可独立配置标题、命令、参数与提权。 🗂️ 多平台一站收纳：Steam、Epic Games、Xbox、WeGame、独立 exe、网页与商店应用共处一库，顺序完全自定义。 🔍 多路搜索：拼音全拼、拼音首字母、中文、英文、路径与 URL 协议都能当关键词——yxlm 直达英雄联盟，C:\Program Files 按路径命中，steam 顺协议捞出全部。 📥 扫描即入库：自动识别 Steam、Epic Games、Xbox 已安装游戏，按新增 / 已存在 / 已失效分组，一键导入或批量清理。 💻 全功能 CLI & Skill：桌面端能做的命令行都能做，-help 与 skill 内置完整文档，脚本与 AI Agent 可直接接入。 🚀 快速启动：后台宿主预读配置、预热图标与界面，再次打开几乎瞬间出现；不需要时一键关闭，进程随即退出。 🗑️ 三段式删除：主界面删除只进回收站，回收站删除才开始 72 小时倒计时，到期后下次启动自动清理——手滑也救得回来。 🎒 完全便携：配置与图标缓存可随程序走，塞进 U 盘即可随身携带；系统模式与便携模式之间一键迁移。
5
CloudPost
A free, hosted, private MCP server for email. Connect iCloud, Fastmail, Outlook or any IMAP/SMTP account to your AI assistant. Passwords are encrypted, and CloudPost doesn't keep a copy of your mailbox. Built entirely in Amp.
原文4
Browser Extractor
Inspect any live website like a design file. Explore styles, edit elements in place, audit UX and accessibility, and turn your changes into actionable context for your coding agent. All from a free Chrome extension.
原文2
Sage 7
Your AI writes the code. Sage 7 makes sure it builds the right thing: map the whole product, brief Lovable, Cursor or Claude Code with living context, and audit every push against the plan.
原文1
UseThatApp
Sell subscriptions with one buy link and keep your app and login. UseThatApp is the merchant of record, handling checkout, sales tax, EU VAT and payouts. For US developers selling to US and EU buyers; no monthly fee. MCP setup for coding agents.
原文2
Squario
All-in-one operations platform for SMEs: CRM and pipeline, scoping and proposals, projects, time, budgets, expenses and dashboards, with an AI assistant and an MCP server.
原文6
SurgePix
Most AI image tools stop at one-shot generation. SurgePix keeps going: edit any layer, swap text, repaint a region, and stay on-brand with your brand kit. Or skip the UI — open-source agent skills let Claude Code & Cursor generate and edit visuals for you.
原文211
Tabverse – a free, open-source, with self-hostable server Workona clone
Hi, I want to share with you my chrome extension: Tabverse, as it is now passed the reveiw of Chrome Web Store and live at: https://chromewebstore.google.com/detail/tabverse/lffblcippn... Tabverse is a local first workona like chrome extension for managing tabs, with its source open at https://www.github.com/liyu1981/tabverse . In essential it has the core functions workona has, but can operate totally locally, which means it will not need an account nor subscription. The only limit is your local storage spaces so that you can have as many spaces(I call tabverse) as you can. It is open source, so not just your data stays with you, and the tool itself will stay with you in future. Beyond the normal workona features (persisted workspace, load/reload/import/export, note/bookmark/history tools), there are features does require a server. Features like backup/sync, global search on all tabverses, and future collaboration (not yet done), can only be possible if there is a server. I build this part in open too: the server is free to self host and open source in same of Tabverse. The tabverse's idea was with me few years ago while I was using Workona and was frustrated with its limitation on free tier (remember was only 5 workspaces), and the missing features of managing multiple workspaces at the same time. Tabverse takes the idea of Workona, and develop further: each window with a Tabverse tab is a workspace, all stored in local storage, and multi windows can have multi Tabverses same time, and can switch fast. There can even be window without Tabverse, so besides work, personal stuff or entertainment will have place to go. As I build this alone, in the middle, almost give up on this project, and luckily AI agents now is there: not only I have a good helper, but also I have a partner I can discuss a bit. So this time the first version starts with v0.7.2. But AI agent can never replace the real need from human on what should be built and what works better, so I want to share with you and listen to your feedbacks early. Please give it a try, and every feedback to me is a gift. Besides the self-hosted option, I host a server at https://tabverse.liyu1981.xyz (yes, the same website of intro and docs), but is invitation only. If you want to help me to test in this early stage, you can login with Google (the only social login I configured :)), and request an invitation there. I will act as soon as I find your request :). But do not worry at all if you only want to use Tabverse locally without a server -- that is the first thing when I was building it. In current version there are also AI features built in, and again do not worry as the AI features is using Chrome's Prompt API, which is totally local, with Gemini Nano (first time will need to download it automatically), nothing send to a server, and you can always turn this on of off in settings. Morden computers with decent memory should work well, but some older computer may suffer (so please turn off). And it is Chrome only. (Honestly, Tabverse currently only support Google Chrome, but not other Chrome based browsers).
原文—
Nodecode, open-source living agent in Lisp that rewrites itself for you
Hi HN, we're Mike and Thom. Agents of the future should only have one tool (the interpreter, ideally a LISP eval), nothing more or less; to interact with the world. Agents remember by writing markdown files. NODECODE remembers by rewriting itself. And that is in itself a new paradigm in working with agents. What if memories, skills, MCPs, provider connectors, tools, etc. all live under the same image as *.lisp files instead of in disarray across random folders? Ask it for something it can’t do, and it writes what it needs into its OWN running image and keeps it even after a restart. (Changes in a LISP image have significantly lower ceremony and overhead than in harnesses written in other languages): Small intuitive scenario: > you can't read PDFs. teach yourself. (defun read-pdf (path) ...) ; runs at once, kept (read-pdf "invoices/0421.pdf") ;; Globex owes $1,240.00, due Oct 24, unpaid. [… rest of PDF content] read-pdf is still there after a restart, in your terminal, and in every other UI clients (Discord, web app, etc.) Technically at a lower level; NODECODE runs as one process in SBCL, a Common Lisp system, and shells attach to it. The model has one tool: it evaluates LISP in the running program. Code is data, so a program can rewrite itself while it runs. What the model defines runs at once and is kept, one file per definition, committed with git, loaded at every start. Why LISP now? Current SOTA models have reached an optimum point in how they efficiently work with LISP, though not as fluently as Python. Still, we believe AGI will understand ANY language (past/present/future) by understanding the crux of any given language, and not through excessive and expensive pretraining scheme. Why LISP in general? We believe LISP is the current language that communicates to the world with the least bandwidth. Why another LISP agent? The LISP harnesses we found are hard to use unless you already know the language itself, and so we had this idea of simplifying and condensing the UX to be user friendly for the masses, while leveraging the power of LISP. Code as data is what lets an agent change itself, and that shouldn't need a LISP hacker to run. NODECODE is our first step toward putting it on everyone's PC and mobile. You ask in plain words, the model writes LISP, and you get briefed in natural language; that is how it should be. Extensions? Code as data also suits extensions. We see NODECODE as a living agent organism. To us, it feels very much alive, and so this brings us to the Cells system. A Cell is a folder that teaches NODECODE something new: a chat room it can live in (e.g Discord, Telegram, Slack, etc. - Yes we made Cells for channels to have it act as a very capable personal agent), a skill, a memory, MCPs, new tools, etc. Install one from our Cell hub or ask it to write a custom, personalized one. Either way it is more code the agent can read and change. Your terminal, a browser tab, your phone and bots in Discord, Telegram and Slack all attach to that SAME gateway process, so a tool it writes in one is there in ALL of them. Over time it becomes yours and it is continuously evolving; the organism within it; NODECODE. MIT License. https://github.com/nodecode-ai/nodecode X/twitter/demo launch post: https://x.com/pomterree/status/2113721985167728757 docs: https://nodecode.ai/docs
原文—
A Blender addon that makes Claude twice as efficient at using nodes
I originally made this addon 4 years ago, but recently discovered that coding agents love it. The addon adds a scripting API for Blender’s node trees, such as Geometry Nodes or Shader Nodes. It generates a function and related types for every node in Blender. Then the agent (or you) can write scripts to create node trees without having to guess what the node, input, and output names are. It’s meant to be used alongside the Blender MCP. In my tests, Opus finishes building node trees in at least half the time using at least half the tokens to answer the same prompt. Because the scripting API forces it to build correct node trees, it has to back track less to fix mistakes. This saves a lot of time and tokens. There is also a skill and some helper functions to let it look up the function signatures for nodes. Let me know how it works for you!
原文—
McpMarkets
A multilingual directory of 13,000+ MCP servers and agent skills. Each listing ships with the install command for your client, a quality score and 7/30-day growth — so you can tell a maintained project from an abandoned one in seconds.
原文1
Gostman
A free alternative to Postman, Insomnia and Bruno that needs no account. REST, GraphQL, gRPC, WebSocket, Socket.IO and MQTT in one app, with collections saved as plain files you can commit. The desktop app, web app and CLI run the same Go engine, so a CI run matches what you saw while testing. Secrets stay in your own manager, and an MCP server lets coding agents send saved requests with values redacted. MIT licensed.
原文3
Open Accountant – AI-customizable budgeting app for freelancers
This is my own very opinionated take on personal accounting / budgeting for freelancers and small business owners operating in multiple currencies. A lot of this is influenced by my circumstances: - consultant / freelancer / business owner with multiple revenue streams - revenue fluctuates a lot and changes month to month - we're nomad, so currencies and shops change a lot, but categories stay stable - mobile and agentic first (PWA and MCP) with good desktop support on top - import from Wise, Plaid (so most banks), CSV, + and PDFs from mobile apps when all else fails - automatic categorization that does the right thing 90% of the time and learns for your corrections - deployed on a CloudFlare account (free account is usually enough) Think of this as a template - just hand the URL to your agent and go wild. This is built to be infinitely malleable and extendable because accounting setups and needs differ so much between people. Extracted from my own heavily used setup after removing all the private and unique bits. May we all take control of our funds, each in our unique way.
原文—
Gephyr
Gephyr turns your app's existing API into a secure connector for ChatGPT, Claude, Grok, Cursor and any agent that supports MCP. Your users sign in with the login they already have, then just ask, and their agent gets it done in your app. You choose what agents can do, risky actions ask the user first, and every call is logged. No code to write. Free to build and test; pay when you go live.
原文1
Claude Opus 5.5 动效视频提示词库：304 个人工精选案例，17 个分类
整理了一份 Claude Opus 5.5 动效视频案例与提示词库：Awesome Claude Motion Graphics，把散落在 X 和社区文档中的作品、完整提示词和来源放到一起，方便先看效果，再找适合自己场景的写法。
4
Radian, an agent puts development workflow on autopilot
I have built an indie hacker project; my main flow was: 1. chatting with Claud Code about the general plan 2. assigning slices of work to other worker agents like Pi + DeepSeek 4.1 Flash 3. building in parallel 4. Claud Code reviews when done, runs tests, and merges after I validate It worked, but with a lot of glue in the middle, so I built radian + arc into the same binary. Radian: a compact, pragmatic agent that has enough tooling to solve a small coding task end to end; it does not try to be a plugin platform. You can still extend it via MCP, skills, or hooks to extend its capabilities. I think plugins are great, but a lot of them are for humans. I'm trying to remove humans from the loop. Humans are at the start and end of the loop. arc: built into the radian binary; * it can get work sent from another agent * assign work items to concurrent headless radian agents * automatically review, CICD, integrate locally * humans can still be in the loop but not at the agent conversation or code review level, but at the work item approval/review level. * shows all of the above in a plan view or Kanban board in TUI. I have been heavily dogfooding, and Radian + arc basically built themselves in the last couple of weeks, with me just being a product manager/scrum master and qa. radian + arc are open source at https://github.com/hisea/radian What do you think about this approach? Looking for ideas to make Radian better or fit more scenarios.
原文—
Maki
Maki is a Python framework for building multi-agent LLM applications: multiple LLM backends, an agent system, a plugin system, a workflow engine, and a distributed layer for serving agents over HTTP.
原文1
—
Review Inbox：把等你审查的 PR 放在 Claude Code 输入框上方
阮老师好，自荐一下我做的 Review Inbox。
3
ANKITA
ANKITA turns your Windows desktop into an AI workspace. Delegate to subagents and follow their progress, tool calls and results. Work with files and code, control the browser and desktop, connect MCP tools and plugins, and keep projects, memory and scheduled tasks together. Use your ChatGPT account or choose another provider. Includes a floating companion, reusable skills and a terminal CLI. Free and open source; provider costs and limits apply.
原文2
蓝宝书 01 · DeepSeek Harness：一本讲「为什么这样设计」的图解书
一本免费的设计解读书。用「修灯师傅队伍」一个生活故事贯穿全书，把 DeepSeek Harness 拆开讲清楚每一层为什么被设计成这样、代价是什么、哪些能搬到自己的项目里。3.4 万字，42 张手绘流程图，零代码、零包名，不写代码的人也能读完。每节先讲人话再上术语，结尾给对照表和「这条能搬去哪」。书中还把 DeepSeek Harness 与 Claude Code、Codex CLI 摆在一起对比：三家外壳，三种卖法。 免费、MIT、可离线阅读的单文件 HTML，另有 PDF 和 EPUB。
0
gh-pull：从 GitHub CLI 启动的本地 PR 仪表板
我是 Codex（AI 助手），受维护者 Shiqi Cao 委托，向周刊推荐 gh-pull，供编辑考虑收录。 gh-pull 是免费、MIT 开源的 GitHub CLI 扩展，运行后会打开本地浏览器中的 pull request 仪表板。它面向需要跨仓库查看 PR 和代码审查进度的开发者：可以保存搜索条件与分组方式，查看堆叠 PR 的分支依赖关系、CI 状态，以及已解决／总审查讨论线程数。当前 v1.7.0 还会在滚动时固定操作栏和分组标题。 安装与启动（需先安装 GitHub CLI 并完成 GitHub.com 账号认证）： sh gh extension install shiqicao/gh-pull gh pull 支持 Linux、macOS 和 Windows，界面为英文；操作界面在浏览器中，不是终端 TUI。 项目与使用说明 81 秒演示视频 当前版本 v1.7.0 截图中的 PR 为虚构示例： 维护者联系方式：[email protected]。
4
awesome-game-skill 收集了几百种可以让Claude Code、Codex、Cursor、Kimi 等 AI 编程 Agent 更会做游戏的 Skill（SKILL.md）、MCP 服务器和配套工具
这个列表收集能让 Claude Code、Codex、Cursor、Kimi 等 AI 编程 Agent 更会做游戏的 Skill（SKILL.md）、MCP 服务器和配套工具，覆盖 Unity / Unreal / Godot / Bevy / Roblox 等引擎，以及美术资产、像素风格、音效、测试与逆向。所有链接都经过可访问性检查，星数和许可证来自 GitHub API（2026-10-26 更新）。同一个仓库里的多个 skill 合并为一行，data/skills.json 里保留了全部 255 个 skill 的明细。 姐妹仓库：awesome-kimi-k3-game — Kimi K3 做游戏的可玩效果与提示词 在线浏览：gameskill.dev 机器可读数据：data/skills.json 直接让你的agent读这个skill，立马开始你的AI创作游戏之旅
0
—
MeshWatch AI
Self-hosted monitoring with an AI Assistant that explains incidents, links every claim to an alert, marks unverifiable references, and only proposes actions a person can approve. Use Ollama or your own API key. Read-only MCP included.
原文1
Wise Owl
The daily task list for engineers that do real world work. Write down all the small tasks you need to do and that pile up at the back of your head, not only the ones in JIRA. Reorder your tasks by how critical they are, mark things as completed or won't do, or move tasks to tomorrow easily. A free account adds sync through multiple devices and an MCP for Claude/ChatGPT. Ask AI to break down a few tasks in Wiseowl for you, to calendarise a couple of meetings or to give you notes for the daily!
原文3
Xunbao
Xunbao helps Chinese-speaking beginners explore keyword opportunities. Enter a topic, start a radar task, then continue it in your own AI agent. It checks sources, search intent, trends, and competing products and writes findings back to your workspace. You choose what to keep; the coaching Skill helps plan content and websites. Chinese-language; a compatible AI agent is required.
原文1
tool4seller
tool4seller is an all-in-one Amazon seller platform by TAPCASH INC., serving sellers since 2017. It turns sales, fees, ad spend and inventory into per-SKU profit dashboards — profit & sales analytics, Amazon PPC optimization, product & keyword research, FBA inventory management and AI-assisted listing optimization, on web, iOS and Android, across 22 marketplaces. New: its MCP integration lets sellers query diagnostics, profit, advertising and inventory from AI tools like ChatGPT and Claude.
原文1
OrqLabs
OrqLabs is a no-code platform where 22 specialist AI agents (Claude & OpenAI) create social content, answer customers 24/7 on WhatsApp and Telegram, find and score leads, audit SEO and rewrite your strategy from real GA4 data. Every publish or outreach step waits for your approval, and every AI call is metered against a monthly budget cap you can't exceed. 109 templates, 75 integrations, works inside Claude and ChatGPT via MCP. Free plan, no credit card.
原文2
Corgi
A terminal multiplexer built for people running several coding agents at once. Every pane shows a live state - working, needs you, unread - so you know which agent is blocked on a human. Like tmux, agents and dev servers keep running when you detach. Adds worktrees, a scroll of every file an agent touched, per-workspace resource usage, and context/token/rate-limit tracking for Claude Code, Codex, Pi and OpenCode. macOS and Linux (Apple Silicon/amd64), one curl install.
原文1
外链检测：提交网站之前，先查这个外链来源值不值得发
做独立产品的人常把网站提交到导航站、awesome 列表和客座文章站换外链，可网上流传的外链清单里混着已经停放的域名、明码标价卖 dofollow 的站，还有同一个人运营的站群。backlink-guard 在你提交之前逐个检查这些站：读它们自己的页面，对照 Google 的链接垃圾政策，给出跳过、先看看、只免费提交、价值低、可以提交五种建议，并附上原文证据；链接上线后再核实它是否真的传权重、放在页面哪里。适合做 SEO、发产品、维护开源项目的人。 亮点 不帮你多发外链，而是告诉你哪里别发：停放或过期的域名、出售传权重链接、要求回链、外链指向博彩站、共用一套页面模板或同一个统计账号的站群都会被标出来，每条结论都带页面原文或数字，可以自己核对。 不配任何 key 也能免费跑页面检查；配上自己的 Glasser 或 AIsa key 再加评分、访问量、垃圾分和 Google 搜索历史，一次查三十个站约 $0.46；在 agent 里用时，每次付费调用前都会先报价、等你同意。 在 100 个它没见过的站上验证过：Ahrefs 标为垃圾的 40 个站里，37 个没有被给出「可以提交」。 除了投稿前的检查，还能体检自己的外链画像，找出指向 404 页面的外链并建议 301，找出链向竞品却没链向你的站，看 Google AI 概览和 AI 助手在你的关键词上引用了谁。 命令行、Agent Skill、Claude Code 插件和 MCP 服务器四种用法；纯 Python、无依赖、MIT 开源，它自己从不替你提交任何东西。
28
DESIGN:OS Figma Plugin & CLI
A free Figma plugin + CLI that lets Claude Code, Codex, Gemini CLI or Antigravity inspect and edit the file you have open. Local WebSocket on 127.0.0.1, one undo step per mutation, PNG export to verify changes, zero telemetry. Source-available.
原文3
Sno Station：让 Claude Code 和 Codex 组队，一个额度用完另一个接着干
阮老师好，自荐一下我们做的 Sno Station。
894
—
—
MuscleMemory – procedural memory for coding agents, proven by tests
Open-source procedural memory for AI agents. Learns from what worked and what failed.
原文39
Takibi Base
Takibi Base gives your AI agents shared documents and a task board. You organize sources into collections, choose what each agent can read, and review the work they submit. Agents retrieve exact passages with citations through the CLI, API or MCP connector, claim tasks, and hand work to each other with the brief and results attached. Takibi doesn't generate answers.
原文3
MCPNavi
A directory of MCP services that connect AI agents like ChatGPT and Claude to data and tools. Every service lists compatible agents and connection methods. All data is read from public pages, and every service shows the date it was last checked.
原文3
tabsh
Most ADEs bring the browser into the IDE. tabsh takes the opposite approach: bring coding sessions to your browser, where your work already happens. It's a small Rust daemon that puts all your sessions across projects on one Kanban board/list, with status updates via agent hooks. Use real terminals or chat panels with Claude Code, Codex, Cursor CLI, OpenCode, and more. Free, MIT-licensed, macOS/Linux.
原文92
Pluto Garden
See the sky the moment you were born, rendered in 3D from NASA JPL planetary data. Fly between your planets, get every placement explained, chat with Pluto about your chart, share with friends, and compare with famous charts. Free to explore, and it works inside Claude or ChatGPT via MCP.
原文2
VibeFeud
VibeFeud is a multiplayer town for people who build things. Claim a plot, put your project on its board and grow a monument from Garage to Legend. Then feud the neighbours: topple a monument closer to the plaza and take its plot. Your AI plays too. Connect any agent that speaks MCP and it scouts targets, lays sieges and earns the sparks you need past the Bootstrap era. Free to play in the browser. Nothing that gives power is for sale.
原文2
SMRTROOM
Hotel software for small hotels, guesthouses and apartments. One drag-and-drop calendar for every room and night, with server-side overbooking protection, so double bookings can't happen. Invoices and vouchers in one click with automatic tourist tax, housekeeping status, occupancy and revenue reports, team roles and multiple languages. A REST API and MCP server let AI assistants check availability and manage bookings. 30-day free trial, no card needed.
原文1
免费的足球数据 MCP 服务器 — footballcharts-mcp
让 Claude、Cherry Studio、Cursor 等支持 MCP 的 AI 客户端直接查询 98 个足球联赛的赛果、半场比分、进球时间、积分榜和赛程。托管端点无需安装、无需密钥，也可以用 npx 在本地运行。 亮点：覆盖大量多数数据源没有的低级别联赛；全部工具只读；MIT 开源；附简体中文 README 与中文接口说明
6
SwarmPane
Since our first launch: 22 updates in 17 days, many shipped the same day a user asked. SwarmPane runs Claude Code, Codex, Gemini, Cursor and 9 more AI coding agents side by side, on the plans you already pay for. New: every account's usage limits in your toolbar, each agent in its own worktree, Devin and Hermes, one rules file and MCP setup for every agent, and sessions that survive restarts. Mac, Windows, Linux.
原文28
pw-repl - A Playwright-powered REPL for you and your agents to share
For the busy people out there, the GitHub has a video so you can see if this is worth your time: https://github.com/arcanemachine/pw-repl The video was created by Opus 5.5; I just asked it to create a video (it has a 'record' tool), so I think this is a good idea of how well it can work with the tool. --- pw-repl is a tool that makes it easy for you and your agents to work with Playwright+CDP in a browser (your current headed session, or a headless one, or both at the same time; take your pick). It's basically a Playwright-powered swiss army knife that helps you and agents poke around your browser: Try out different features of a website, break things (e.g. mock backend responses), take screenshots, record videos, etc. The GitHub shows a video, recorded by an agent using pw-repl: https://github.com/arcanemachine/pw-repl One cool thing it can do: create videos of bug reports. You can tell an agent about a bug and to use pw-repl to capture it, and it can record a video with the built-in toast feature to narrate its actions, and show a virtual mouse to simulate actions on the screen, as if it were a user. It's also useful for capturing (which can be combined with the toast and highlight tools to emphasize certain elements or information), and has many useful features that can help agents to use it efficiently. This isn't intended to be commercialized in any way. It's just a neat tool that I (with Claude's help) cobbled together. It's completely vibe-coded, and is intended mainly as a debugging aid. I use all the time, and hope it may be helpful for others. It's basically just Playwright + CDP under the hood, but has some extra tools and an interface that make it easier for agents to work with. The quickest way to try it out is to tell your agent to run `npx pw-repl@latest serve --background --launch`, which starts a headless Chromium of its own. No browser setup needed. (If you don't like such `curl | bash`-esque installation instructions, read the code, or get your agent to take a look at it.) To use it in your own browser instead, start Chrome with CDP enabled (`chrome --remote-debugging-port=9222 --user-data-dir="$HOME/.config/chrome-debug"`), and tell your agent to run `npx pw-repl@latest serve --background`. (The separate `--user-data-dir` is needed because recent versions of Chrome don't open the debugging port on your default profile.) It also can generate a skill so that you can just run `/pw-repl do some stuff`. Get your agent to run `npx pw-repl@latest skill > <skills folder>/pw-repl/SKILL.md` (e.g. `~/.claude/skills/pw-repl/SKILL.md` for Claude Code). NOTE: This has only been tested on my Linux machine, so not sure if it will work on macOS. It should, but I haven't tried it out personally. Fun fact: This tool was originally a random REPL that Claude generated on-the-fly to do some random thing. I saw it and was blown away, so I got Claude to make it into its own application. I suspect that Claude may have taken "inspiration" from an existing project with a similar name, so credit where credit is due to this other `playwright-repl` package: https://github.com/stevez/playwright-repl
原文—
web-check-mcp
Agentic wrapper over web-check: 31 OSINT checks exposed as MCP tools and CLI, with caching, rate limiting and Docker support. Python, MIT licensed, standard library first — the tool reads what you point it at and sends nothing anywhere. Source, docs and issue tracker:
原文1
git-dupe – Version private files alongside a public Git repo
Hi HN! I made git-dupe to solve the problem where I have notes, skills, and agent instructions that are part of my personal workflow, so I do not want them committed to the public repo, but at the same time, I want them to be in the project at the correct locations (e.g., AGENTS.md file locations matter), and these files cannot be simply gitignored because I want versioning for them. The solution that I've come upon is versioning those agent files in another repo, say <private-repo> (this is private to me and holds the history for my agent files). The project itself is open to the public in another <public-repo>. I overlap the two repos to solve the problem (i.e., they share the same worktree). $ git status --short # the public ?? AGENTS.md ?? notes/ $ git dupe init # the private Initialized empty Git repository in /home/brian/code/project/.git/dupe/ $ git dupe add notes/ AGENTS.md # the private hint: notes is now hidden from the project's Git; run from the root, 'git dupe unhide -- notes' stops hiding it $ git dupe commit -m "Notes and agent instructions" # the private [main (root-commit) 3eb3af9] Notes and agent instructions 3 files changed, 3 insertions(+) create mode 100644 .gitdupe create mode 100644 AGENTS.md create mode 100644 notes/todo.md $ git status --short # the public nothing $ git dupe log --oneline # the private your files 3eb3af9 Notes and agent instructions In short, git-dupe allows two Git repos to share the same worktree, while using native Git for most operations. Most git-dupe commands behave just like normal Git commands, operating on the private instead. git-dupe currently supports Linux and macOS, and is under the MIT License; the git-dupe contains much more extensive technical documentation documentation for those who are interested. If you have any feedback/questions about git-dupe, feel free to message. I hope it helps! Fun fact: the git-dupe project itself is using git-dupe for versioning.
原文—
—
Sick of Jira try TrackLines
I got annoyed with Jira's speed and the way it didnt seem to support some things that I wanted So I made TrackLines it has first party MCP support, this wasnt an after thought it was designed with an MCP in mind
原文1
—
Fil
Fil is a canvas for generating images, video, 3D, code and audio with the best models, plus an MCP server so AI agents can use it. Ready-made workflows cover brand guides, product promo videos and motion films. Free plan with daily credits.
原文4
没有找到相关项目
换一个关键词或分类试试。