[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"similar-ezyang--codemcp":3,"tool-ezyang--codemcp":64},[4,17,27,35,48,56],{"id":5,"name":6,"github_repo":7,"description_zh":8,"stars":9,"difficulty_score":10,"last_commit_at":11,"category_tags":12,"status":16},3808,"stable-diffusion-webui","AUTOMATIC1111\u002Fstable-diffusion-webui","stable-diffusion-webui 是一个基于 Gradio 构建的网页版操作界面，旨在让用户能够轻松地在本地运行和使用强大的 Stable Diffusion 图像生成模型。它解决了原始模型依赖命令行、操作门槛高且功能分散的痛点，将复杂的 AI 绘图流程整合进一个直观易用的图形化平台。\n\n无论是希望快速上手的普通创作者、需要精细控制画面细节的设计师，还是想要深入探索模型潜力的开发者与研究人员，都能从中获益。其核心亮点在于极高的功能丰富度：不仅支持文生图、图生图、局部重绘（Inpainting）和外绘（Outpainting）等基础模式，还独创了注意力机制调整、提示词矩阵、负向提示词以及“高清修复”等高级功能。此外，它内置了 GFPGAN 和 CodeFormer 等人脸修复工具，支持多种神经网络放大算法，并允许用户通过插件系统无限扩展能力。即使是显存有限的设备，stable-diffusion-webui 也提供了相应的优化选项，让高质量的 AI 艺术创作变得触手可及。",162132,3,"2026-04-05T11:01:52",[13,14,15],"开发框架","图像","Agent","ready",{"id":18,"name":19,"github_repo":20,"description_zh":21,"stars":22,"difficulty_score":23,"last_commit_at":24,"category_tags":25,"status":16},1381,"everything-claude-code","affaan-m\u002Feverything-claude-code","everything-claude-code 是一套专为 AI 编程助手（如 Claude Code、Codex、Cursor 等）打造的高性能优化系统。它不仅仅是一组配置文件，而是一个经过长期实战打磨的完整框架，旨在解决 AI 代理在实际开发中面临的效率低下、记忆丢失、安全隐患及缺乏持续学习能力等核心痛点。\n\n通过引入技能模块化、直觉增强、记忆持久化机制以及内置的安全扫描功能，everything-claude-code 能显著提升 AI 在复杂任务中的表现，帮助开发者构建更稳定、更智能的生产级 AI 代理。其独特的“研究优先”开发理念和针对 Token 消耗的优化策略，使得模型响应更快、成本更低，同时有效防御潜在的攻击向量。\n\n这套工具特别适合软件开发者、AI 研究人员以及希望深度定制 AI 工作流的技术团队使用。无论您是在构建大型代码库，还是需要 AI 协助进行安全审计与自动化测试，everything-claude-code 都能提供强大的底层支持。作为一个曾荣获 Anthropic 黑客大奖的开源项目，它融合了多语言支持与丰富的实战钩子（hooks），让 AI 真正成长为懂上",138956,2,"2026-04-05T11:33:21",[13,15,26],"语言模型",{"id":28,"name":29,"github_repo":30,"description_zh":31,"stars":32,"difficulty_score":23,"last_commit_at":33,"category_tags":34,"status":16},2271,"ComfyUI","Comfy-Org\u002FComfyUI","ComfyUI 是一款功能强大且高度模块化的视觉 AI 引擎，专为设计和执行复杂的 Stable Diffusion 图像生成流程而打造。它摒弃了传统的代码编写模式，采用直观的节点式流程图界面，让用户通过连接不同的功能模块即可构建个性化的生成管线。\n\n这一设计巧妙解决了高级 AI 绘图工作流配置复杂、灵活性不足的痛点。用户无需具备编程背景，也能自由组合模型、调整参数并实时预览效果，轻松实现从基础文生图到多步骤高清修复等各类复杂任务。ComfyUI 拥有极佳的兼容性，不仅支持 Windows、macOS 和 Linux 全平台，还广泛适配 NVIDIA、AMD、Intel 及苹果 Silicon 等多种硬件架构，并率先支持 SDXL、Flux、SD3 等前沿模型。\n\n无论是希望深入探索算法潜力的研究人员和开发者，还是追求极致创作自由度的设计师与资深 AI 绘画爱好者，ComfyUI 都能提供强大的支持。其独特的模块化架构允许社区不断扩展新功能，使其成为当前最灵活、生态最丰富的开源扩散模型工具之一，帮助用户将创意高效转化为现实。",107662,"2026-04-03T11:11:01",[13,14,15],{"id":36,"name":37,"github_repo":38,"description_zh":39,"stars":40,"difficulty_score":23,"last_commit_at":41,"category_tags":42,"status":16},2268,"ML-For-Beginners","microsoft\u002FML-For-Beginners","ML-For-Beginners 是由微软推出的一套系统化机器学习入门课程，旨在帮助零基础用户轻松掌握经典机器学习知识。这套课程将学习路径规划为 12 周，包含 26 节精炼课程和 52 道配套测验，内容涵盖从基础概念到实际应用的完整流程，有效解决了初学者面对庞大知识体系时无从下手、缺乏结构化指导的痛点。\n\n无论是希望转型的开发者、需要补充算法背景的研究人员，还是对人工智能充满好奇的普通爱好者，都能从中受益。课程不仅提供了清晰的理论讲解，还强调动手实践，让用户在循序渐进中建立扎实的技能基础。其独特的亮点在于强大的多语言支持，通过自动化机制提供了包括简体中文在内的 50 多种语言版本，极大地降低了全球不同背景用户的学习门槛。此外，项目采用开源协作模式，社区活跃且内容持续更新，确保学习者能获取前沿且准确的技术资讯。如果你正寻找一条清晰、友好且专业的机器学习入门之路，ML-For-Beginners 将是理想的起点。",84991,"2026-04-05T10:45:23",[14,43,44,45,15,46,26,13,47],"数据工具","视频","插件","其他","音频",{"id":49,"name":50,"github_repo":51,"description_zh":52,"stars":53,"difficulty_score":10,"last_commit_at":54,"category_tags":55,"status":16},3128,"ragflow","infiniflow\u002Fragflow","RAGFlow 是一款领先的开源检索增强生成（RAG）引擎，旨在为大语言模型构建更精准、可靠的上下文层。它巧妙地将前沿的 RAG 技术与智能体（Agent）能力相结合，不仅支持从各类文档中高效提取知识，还能让模型基于这些知识进行逻辑推理和任务执行。\n\n在大模型应用中，幻觉问题和知识滞后是常见痛点。RAGFlow 通过深度解析复杂文档结构（如表格、图表及混合排版），显著提升了信息检索的准确度，从而有效减少模型“胡编乱造”的现象，确保回答既有据可依又具备时效性。其内置的智能体机制更进一步，使系统不仅能回答问题，还能自主规划步骤解决复杂问题。\n\n这款工具特别适合开发者、企业技术团队以及 AI 研究人员使用。无论是希望快速搭建私有知识库问答系统，还是致力于探索大模型在垂直领域落地的创新者，都能从中受益。RAGFlow 提供了可视化的工作流编排界面和灵活的 API 接口，既降低了非算法背景用户的上手门槛，也满足了专业开发者对系统深度定制的需求。作为基于 Apache 2.0 协议开源的项目，它正成为连接通用大模型与行业专有知识之间的重要桥梁。",77062,"2026-04-04T04:44:48",[15,14,13,26,46],{"id":57,"name":58,"github_repo":59,"description_zh":60,"stars":61,"difficulty_score":10,"last_commit_at":62,"category_tags":63,"status":16},2181,"OpenHands","OpenHands\u002FOpenHands","OpenHands 是一个专注于 AI 驱动开发的开源平台，旨在让智能体（Agent）像人类开发者一样理解、编写和调试代码。它解决了传统编程中重复性劳动多、环境配置复杂以及人机协作效率低等痛点，通过自动化流程显著提升开发速度。\n\n无论是希望提升编码效率的软件工程师、探索智能体技术的研究人员，还是需要快速原型验证的技术团队，都能从中受益。OpenHands 提供了灵活多样的使用方式：既可以通过命令行（CLI）或本地图形界面在个人电脑上轻松上手，体验类似 Devin 的流畅交互；也能利用其强大的 Python SDK 自定义智能体逻辑，甚至在云端大规模部署上千个智能体并行工作。\n\n其核心技术亮点在于模块化的软件智能体 SDK，这不仅构成了平台的引擎，还支持高度可组合的开发模式。此外，OpenHands 在 SWE-bench 基准测试中取得了 77.6% 的优异成绩，证明了其解决真实世界软件工程问题的能力。平台还具备完善的企业级功能，支持与 Slack、Jira 等工具集成，并提供细粒度的权限管理，适合从个人开发者到大型企业的各类用户场景。",70612,"2026-04-05T11:12:22",[26,15,13,45],{"id":65,"github_repo":66,"name":67,"description_en":68,"description_zh":69,"ai_summary_zh":69,"readme_en":70,"readme_zh":71,"quickstart_zh":72,"use_case_zh":73,"hero_image_url":74,"owner_login":75,"owner_name":76,"owner_avatar_url":77,"owner_bio":78,"owner_company":78,"owner_location":79,"owner_email":80,"owner_twitter":78,"owner_website":81,"owner_url":82,"languages":83,"stars":92,"forks":93,"last_commit_at":94,"license":95,"difficulty_score":10,"env_os":96,"env_gpu":97,"env_ram":97,"env_deps":98,"category_tags":103,"github_topics":78,"view_count":23,"oss_zip_url":78,"oss_zip_packed_at":78,"status":16,"created_at":104,"updated_at":105,"faqs":106,"releases":142},2010,"ezyang\u002Fcodemcp","codemcp","Coding assistant MCP for Claude Desktop","codemcp 是一个让 Claude Desktop 变成你本地代码库的编程搭档的开源工具。你无需复制粘贴代码，可以直接让 Claude 在你的项目中修改文件、修复 Bug、重构代码，甚至运行测试，所有操作都在本地完成，像一位懂代码的同事在你身边协作。\n\n它解决了 AI 编程助手常需反复粘贴代码、缺乏版本控制、操作不可逆的问题。codemcp 默认自动接受修改，但会把每一次 AI 的改动自动提交到 Git，你可以随时回滚到任意版本；同时，它禁止 AI 随意执行 Shell 命令，必须提前在配置文件中声明允许的指令，保障安全。此外，它不绑定任何 IDE，无论你用 Vim、VSCode 还是其他编辑器，都能无缝配合。\n\n适合习惯命令行、重视代码版本控制、偏好使用 Claude Pro 的开发者。虽然 Anthropic 已推出官方 Claude Code，但 codemcp 保留了更灵活的配置和开源自由，尤其适合希望掌控 AI 行为边界、不喜欢封闭生态的用户。其 Git 版本化修改机制，是目前多数 AI 编程工具中少见的实用设计。","NOTICE: claude code is available with Anthropic's $20\u002Fmo subscription, so I consider\ncodemcp fully obsolete.  However, there are some good design ideas (especially around\nthe Git-versioning scheme) that I eventually want to port into the current generation\nof agentic coding clis.\n\n# codemcp\n\nMake Claude Desktop a pair programming assistant by installing codemcp.  With\nit, you can directly ask Claude to implement features, fix bugs and do\nrefactors on a codebase on your computer; Claude will directly edit files and\nrun tests.  Say goodbye to copying code in and out of Claude's chat window!\n\n![Screenshot of Claude Desktop with codemcp](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fezyang_codemcp_readme_595410fcb482.png)\n\ncodemcp offers similar functionality to other AI coding software (Claude Code,\nCursor, Cline, Aider), but it occupies a unique point in the design space:\n\n1. It's intended to be used with **Claude Pro**, Anthropic's $20\u002Fmo\n   subscription offering.  I like paying for my usage with a subscription plan\n   because it means **zero marginal cost** for agent actions; no more feeling\n   bad that you wasted five bucks on a changeset that doesn't work.\n\n   Note that if you have Claude Max ($100\u002Fmo), Claude Code can also be used\n   with subscription based pricing.  The value proposition for codemcp is\n   murkier in this case (and it is definitely inferior to Claude Code in some\n   respects), but you can still use codemcp with Claude Max if you prefer some\n   of the other UI decisions it makes.  (Also, it's open source, so you can\n   change it if you don't like it, unlike Claude Code!)\n\n2. It's built around **auto-accept by default**.  I want my agent to get as\n   far as it can without my supervision, so I can review everything in one go at\n   the end.  There are two key things that codemcp does differently than most\n   coding agents: we **forbid unrestricted shell**, instead requiring you to\n   predeclare commands the agent can use in ``codemcp.toml``, and we **Git\n   version all LLM edits**, so you can roll back agent changes on a\n   fine-grained basis and don't have to worry about forgetting to commit\n   changes.\n\n3. It's **IDE agnostic**: you ask Claude to make changes, it makes them, and\n   then you can use your favorite IDE setup to review the changes and make\n   further edits.  I use vim as my daily driver editor, and coding environments\n   that require VSCode or a specific editor are a turn off for me.\n\n## IMPORTANT: For master users - Major changes for token efficiency\n\nTo improve codemcp's token efficiency, on master I am in the process of\nchanging codemcp back into a multi-tool tool (instead of a single tool whose\ninstructions are blatted into chat when you InitProject).  This means you have\nto manually approve tool use.  Because tool use approval is persistent across\nmultiple chats, I think this is a reasonable tradeoff to make, but if you\nreally don't like, file a bug at\n[refined-claude](https:\u002F\u002Fgithub.com\u002Fezyang\u002Frefined-claude\u002Fissues) browser\nextension for supporting auto-approve tool use.\n\n## Installation\n\nI recommend this specific way of installing and using codemcp:\n\n1. Install `uv` and install git, if they are not installed already.\n\n2. Install [claude-mcp](https:\u002F\u002Fchromewebstore.google.com\u002Fdetail\u002Fmcp-for-claudeai\u002Fjbdhaamjibfahpekpnjeikanebpdpfpb) on your browser.\n   This enables you to connect to SSE MCP servers directly from the website,\n   which means you don't need to use Claude Desktop and can easily have\n   multiple chat windows going in parallel.  We expect this extension should\n   be soon obsoleted by the rollout of\n   [Integrations](https:\u002F\u002Fwww.anthropic.com\u002Fnews\u002Fintegrations).  At time of\n   writing, however, Integrations have not yet arrived for Claude Pro subscribers.\n\n3. Run codemcp using ``uvx --from git+https:\u002F\u002Fgithub.com\u002Fezyang\u002Fcodemcp@prod codemcp serve``.\n   You can add ``--port 1234`` if you need it to listen on a non-standard port.\n\n   Pro tip: if you like to live dangerously, you can change `prod` to `main`.  If\n   you want to pin to a specific release, replace it with `0.3.0` or similar.\n\n   Pro tip: you can run codemcp remotely!  If you use\n   [Tailscale](https:\u002F\u002Ftailscale.com\u002F) and trust all devices on your Tailnet,\n   you can do this securely by passing ``--host 100.101.102.103`` (replace the\n   IP with the Tailscale IP address of your node.  This IP typically lives in\n   the 100.64.0.0\u002F10 range.)  **WARNING:** Anyone with access to this MCP can perform\n   arbitrary code execution on your computer, it is **EXTREMELY** unlikely you want to\n   bind to 0.0.0.0.\n\n4. Configure claude-mcp with URL: ``http:\u002F\u002F127.0.0.1:8000\u002Fsse`` (replace the port if needed.)\n\n5. Unfortunately, the web UI inconsistently displays the hammer icon.  However, you can verify\n   that the MCP server is working by looking for \"[MCP codemcp] SSE connection opened\" in the\n   Console, or by asking Claude what tools it has available (it should say\n   tools from codemcp are available.)\n\nIf you prefer to use Claude Desktop or have unusual needs, check out [INSTALL.md](INSTALL.md) for\ninstallation instructions for a variety of non-standard situations.\n\n## Usage\n\nFirst, you must create a `codemcp.toml` file in the Git repository checkout\nyou want to work on.  If you want the agent to be able to do things like run\nyour formatter or run tests, add the commands to execute them in the commands\nsection (note: these commands need to appropriately setup any virtual\nenvironment they need):\n\n```toml\nformat = [\".\u002Frun_format.sh\"]\ntest = [\".\u002Frun_test.sh\"]\n```\n\nThe ``format`` command is special; it is always run after every file edit.\n\nNext, in Claude Desktop, we recommend creating a Project and putting this in\nthe Project Instructions:\n\n```\nInitialize codemcp with $PROJECT_DIR\n```\n\nWhere `$PROJECT_DIR` is the path to the project you want to work on.\n\nThen chat with Claude about what changes you want to make to the project.\nEvery time codemcp makes a change to your code, it will generate a commit.\n\nTo see some sample transcripts using this tool, check out:\n\n- [Implement a new feature](https:\u002F\u002Fclaude.ai\u002Fshare\u002Fa229d291-6800-4cb8-a0df-896a47602ca0)\n- [Fix failing tests](https:\u002F\u002Fclaude.ai\u002Fshare\u002F2b7161ef-5683-4261-ad45-fabc3708f950)\n- [Do a refactor](https:\u002F\u002Fclaude.ai\u002Fshare\u002Ff005b43c-a657-43e5-ad9f-4714a5cd746f)\n\ncodemcp will generate a commit per chat and amend it as it is working on your feature.\n\n## Philosophy\n\n- When you get rate limited, take the time to do something else (review\n  Claude's code, review someone else's code, make plans, do some meetings)\n\n- This is *not* an autonomous agent.  At minimum, you have to intervene after\n  every chat to review the changes and request the next change.  While you\n  *can* ask for a long list of things to be done in a single chat, you will\n  likely hit Claude Desktop's output limit and have to manually \"continue\" the\n  agent anyway.  Embrace it, and use the interruptions to make sure Claude is\n  doing the right thing.\n\n- When Claude goes off the rails, it costs you time rather than dollars.\n  Behave accordingly: if time is the bottleneck, watch Claude's incremental\n  output carefully.\n\n## Configuration\n\nHere are all the config options supported by `codemcp.toml`:\n\n```toml\nproject_prompt = \"\"\"\nBefore beginning work on this feature, write a short haiku.  Do this only once.\n\"\"\"\n\n[commands]\nformat = [\".\u002Frun_format.sh\"]\ntest = [\".\u002Frun_test.sh\"]\n```\n\nThe `project_prompt` will be loaded when you initialize the project in chats.\n\nThe `commands` section allows you to configure commands for specific tools.  The\nnames are told to the LLM, who will decide when it wants to run them.  You can add\ninstructions how to use tools in the `project_prompt`; we also support a more verbose\nsyntax where you can give specific instructions on a tool-by-tool basis:\n\n```\n[commands.test]\ncommand = [\".\u002Frun_test.sh\"]\ndoc = \"Accepts a pytest-style test selector as an argument to run a specific test.\"\n```\n\n## Troubleshooting\n\nTo run the server with inspector, use:\n\n```\nPYTHONPATH=. mcp dev codemcp\u002F__main__.py\n```\n\nLogs are written to `~\u002F.codemcp\u002Fcodemcp.log`. The log level can be set in a global configuration file at `~\u002F.codemcprc`:\n\n```toml\n[logger]\nverbosity = \"INFO\"  # Can be DEBUG, INFO, WARNING, ERROR, or CRITICAL\n```\n\nLogging is not configurable on a per project basis, but this shouldn't matter\nmuch because it's difficult to use Claude Desktop in parallel on multiple\nprojects anyway.\n\n## Contributing\n\nSee [CONTRIBUTING.md](CONTRIBUTING.md).\n","注意：Claude Code 需要订阅 Anthropic 的每月 20 美元套餐才能使用，因此我认为\ncodemcp 已经完全过时了。不过，codemcp 中有一些不错的设计理念（尤其是围绕\nGit 版本控制方案的），我最终希望将这些理念移植到当前一代的智能编码 CLI 中。\n\n# codemcp\n\n通过安装 codemcp，让 Claude Desktop 成为你的配对编程助手。借助它，你可以直接让 Claude 在你电脑上的代码库中实现功能、修复 bug 和进行重构；Claude 会直接编辑文件并运行测试。从此告别在 Claude 的聊天窗口里复制粘贴代码啦！\n\n![Claude Desktop 与 codemcp 的截图](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fezyang_codemcp_readme_595410fcb482.png)\n\ncodemcp 提供的功能与其他 AI 编码软件（Claude Code、Cursor、Cline、Aider）类似，但它在设计上独树一帜：\n\n1. 它专为 **Claude Pro** 设计，这是 Anthropic 提供的每月 20 美元订阅服务。我喜欢用订阅模式支付使用费，因为这意味着代理操作的**边际成本为零**；再也不用为浪费了五美元却没用上的变更集而感到内疚了。\n\n需要注意的是，如果你使用的是 Claude Max（每月 100 美元），Claude Code 同样可以配合订阅定价使用。在这种情况下，codemcp 的价值主张就显得不那么清晰了（而且某些方面确实不如 Claude Code），但如果你喜欢 codemcp 做出的一些 UI 决策，仍然可以继续使用它。另外，codemcp 是开源的，所以如果你不喜欢它的设计，还可以自己动手修改呢，这可是 Claude Code 所不具备的优势！\n\n2. 它默认采用**自动接受模式**。我希望我的代理能在无需我监督的情况下尽可能地完成任务，这样最后我可以一次性审查所有内容。codemcp 与大多数编码代理的不同之处在于两点：我们**禁止无限制的 shell 操作**，而是要求你在 ``codemcp.toml`` 文件中预先声明代理可以使用的命令；此外，我们还会对大语言模型的所有编辑内容进行**Git 版本控制**，这样你就能以细粒度的方式回滚代理的更改，再也不用担心忘记提交更改了。\n\n3. 它是**与 IDE 无关的**：你让 Claude 做改动，它就直接完成，然后你可以用自己最喜欢的 IDE 环境来审查改动并进一步编辑。我日常习惯用 vim 编辑器，那些要求必须用 VSCode 或特定编辑器的编码环境对我来说简直无法忍受。\n\n## 重要提示：针对高级用户——Token 效率的重大改进\n\n为了提升 codemcp 的 Token 效率，在 master 分支上，我正逐步将 codemcp 改回一个多工具模式（而不是单个工具，每次 InitProject 时都会把指令一股脑扔进聊天中）。这意味着你需要手动批准工具的使用。由于工具使用批准会在多个聊天中保持持久，我觉得这是一个合理的权衡，但如果你实在不喜欢，可以在\n[refined-claude](https:\u002F\u002Fgithub.com\u002Fezyang\u002Frefined-claude\u002Fissues) 浏览器扩展中提交问题，请求支持自动批准工具使用。\n\n## 安装\n\n我推荐以下这种特定的安装和使用方式：\n\n1. 如果尚未安装 `uv` 和 git，请先安装它们。\n\n2. 在浏览器中安装 [claude-mcp](https:\u002F\u002Fchromewebstore.google.com\u002Fdetail\u002Fmcp-for-claudeai\u002Fjbdhaamjibfahpekpnjeikanebpdpfpb)。这能让你直接从网站连接到 SSE MCP 服务器，意味着你无需使用 Claude Desktop，还能轻松同时开启多个聊天窗口。我们预计这个扩展很快会被\n[Integrations](https:\u002F\u002Fwww.anthropic.com\u002Fnews\u002Fintegrations) 的推出所取代。不过，截至撰写本文时，Integrations 尚未面向 Claude Pro 订阅用户开放。\n\n3. 使用 ``uvx --from git+https:\u002F\u002Fgithub.com\u002Fezyang\u002Fcodemcp@prod codemcp serve`` 运行 codemcp。如果需要监听非标准端口，可以加上 ``--port 1234``。\n\n小贴士：如果你喜欢冒险，可以将 `prod` 改为 `main`。如果你想固定到某个具体版本，可以用 `0.3.0` 或其他类似版本号替换。\n\n小贴士：你甚至可以远程运行 codemcp！如果你使用\n[Tailscale](https:\u002F\u002Ftailscale.com\u002F) 并信任 Tailnet 上的所有设备，可以通过传递 ``--host 100.101.102.103`` 来安全地运行（请将 IP 替换为你节点的 Tailscale IP 地址。这个 IP 通常位于 100.64.0.0\u002F10 网段内）。**警告：**任何拥有此 MCP 访问权限的人都能在你的电脑上执行任意代码，因此**极不建议**你绑定到 0.0.0.0。\n\n4. 使用 URL 配置 claude-mcp：``http:\u002F\u002F127.0.0.1:8000\u002Fsse``（如有需要可更换端口）。\n\n5. 不幸的是，网页界面有时会不一致地显示锤子图标。不过，你可以通过查看控制台中的 “[MCP codemcp] SSE 连接已打开”来验证 MCP 服务器是否正常工作，或者直接问 Claude 它有哪些可用工具（应该会显示 codemcp 提供的工具）。\n\n如果你更喜欢使用 Claude Desktop 或有特殊需求，可以查看 [INSTALL.md](INSTALL.md) 获取各种非标准情况下的安装说明。\n\n## 使用方法\n\n首先，你需要在想要工作的 Git 仓库检出目录下创建一个 `codemcp.toml` 文件。如果你想让代理能够运行格式化工具或测试，就在 commands 部分添加相应的命令（注意：这些命令需要正确配置它们所需的虚拟环境）：\n\n```toml\nformat = [\".\u002Frun_format.sh\"]\ntest = [\".\u002Frun_test.sh\"]\n```\n\n其中，``format`` 命令是特殊的，每次编辑文件后都会自动运行。\n\n接下来，在 Claude Desktop 中，我们建议创建一个 Project，并在 Project Instructions 中加入以下内容：\n\n```\n使用 $PROJECT_DIR 初始化 codemcp\n```\n\n其中 `$PROJECT_DIR` 是你要工作的项目的路径。\n\n然后，与 Claude 聊天，告诉它你想对项目做哪些改动。每次 codemcp 对你的代码做出改动时，都会生成一次提交。\n\n想看一些使用该工具的示例对话，可以参考：\n\n- [实现新功能](https:\u002F\u002Fclaude.ai\u002Fshare\u002Fa229d291-6800-4cb8-a0df-896a47602ca0)\n- [修复失败的测试](https:\u002F\u002Fclaude.ai\u002Fshare\u002F2b7161ef-5683-4261-ad45-fabc3708f950)\n- [进行重构](https:\u002F\u002Fclaude.ai\u002Fshare\u002Ff005b43c-a657-43e5-ad9f-4714a5cd746f)\n\ncodemcp 会为每次聊天生成一次提交，并在处理你的功能时不断更新提交内容。\n\n## 哲学\n\n- 当你遇到速率限制时，不妨利用这段时间做些别的事情（比如 review Claude 的代码、review 别人的代码、制定计划、参加一些会议）。\n\n- 这并不是一个自主代理。至少，在每次聊天结束后，你都必须介入，审查所做的更改并请求下一步的修改。虽然你可以在一次聊天中一次性提出一大堆待办事项，但很可能会达到 Claude Desktop 的输出上限，最终仍需手动“继续”代理。请坦然接受这一点，并利用这些中断时刻确保 Claude 正在做正确的事情。\n\n- 当 Claude 开始跑偏时，耗费的是时间而非金钱。因此，请相应地调整你的行为：如果时间成了瓶颈，那就仔细观察 Claude 的逐步输出。\n\n## 配置\n\n以下是 `codemcp.toml` 支持的所有配置选项：\n\n```toml\nproject_prompt = \"\"\"\n在开始这项功能的工作之前，请先写一首简短的俳句。此步骤只需执行一次。\n\"\"\"\n\n[commands]\nformat = [\".\u002Frun_format.sh\"]\ntest = [\".\u002Frun_test.sh\"]\n```\n\n当你在聊天中初始化项目时，`project_prompt` 将会被加载。\n\n`commands` 部分允许你为特定工具配置命令。这些命令的名称会告知大语言模型，由它决定何时运行。你可以在 `project_prompt` 中添加如何使用工具的说明；我们还支持一种更详细的语法，让你能够针对每个工具单独给出具体指令：\n\n```\n[commands.test]\ncommand = [\".\u002Frun_test.sh\"]\ndoc = \"接受一个 pytest 样式的测试选择器作为参数，以运行特定的测试。\"\n```\n\n## 故障排除\n\n要以带调试器的方式运行服务器，请使用：\n\n```\nPYTHONPATH=. mcp dev codemcp\u002F__main__.py\n```\n\n日志会写入 `~\u002F.codemcp\u002Fcodemcp.log`。日志级别可在全局配置文件 `~\u002F.codemcprc` 中设置：\n\n```toml\n[logger]\nverbosity = \"INFO\"  # 可选值：DEBUG、INFO、WARNING、ERROR 或 CRITICAL\n```\n\n日志记录无法按项目单独配置，但这其实并不重要，因为无论如何，很难在多个项目上并行使用 Claude Desktop。\n\n## 贡献\n\n请参阅 [CONTRIBUTING.md](CONTRIBUTING.md)。","# codemcp 快速上手指南\n\n## 环境准备\n\n- **系统要求**：macOS \u002F Linux \u002F Windows（WSL2 推荐）\n- **前置依赖**：\n  - `git`（版本控制）\n  - `uv`（Python 包管理器，推荐替代 pip）\n  - Claude Pro 订阅（$20\u002F月）或 Claude Max（$100\u002F月）\n  - 浏览器安装 [claude-mcp 扩展](https:\u002F\u002Fchromewebstore.google.com\u002Fdetail\u002Fmcp-for-claudeai\u002Fjbdhaamjibfahpekpnjeikanebpdpfpb)\n\n> 注：推荐使用 `uv` 替代 pip，安装更快且兼容性更好。国内用户可使用清华源加速：`pip install -i https:\u002F\u002Fpypi.tuna.tsinghua.edu.cn\u002Fsimple uv`\n\n## 安装步骤\n\n1. 安装 `uv` 和 `git`（如未安装）：\n\n```bash\npip install -i https:\u002F\u002Fpypi.tuna.tsinghua.edu.cn\u002Fsimple uv\n```\n\n2. 启动 codemcp 服务（推荐生产版）：\n\n```bash\nuvx --from git+https:\u002F\u002Fgithub.com\u002Fezyang\u002Fcodemcp@prod codemcp serve\n```\n\n> 可选：指定端口 `--port 1234`，或使用 Tailscale 远程访问（需配置 `--host 100.101.102.103`）\n\n3. 在浏览器中配置 claude-mcp 扩展：\n   - URL 设置为：`http:\u002F\u002F127.0.0.1:8000\u002Fsse`（端口按实际调整）\n\n4. 验证服务：在 Claude 聊天窗口中输入 “What tools are available?”，应看到 codemcp 工具列表。\n\n## 基本使用\n\n1. 在目标项目根目录创建 `codemcp.toml`：\n\n```toml\n[commands]\nformat = [\".\u002Frun_format.sh\"]\ntest = [\".\u002Frun_test.sh\"]\n```\n\n> `format` 命令会在每次文件修改后自动执行。\n\n2. 在 Claude Desktop 中新建 Project，填写项目指令：\n\n```\nInitialize codemcp with $PROJECT_DIR\n```\n\n> 将 `$PROJECT_DIR` 替换为你的项目路径，如 `\u002Fhome\u002Fuser\u002Fmy-project`\n\n3. 开始对话，例如：\n\n> “请为这个项目添加一个登录功能，使用 FastAPI。”\n\ncodemcp 将自动修改文件、运行测试，并为每次变更生成 Git 提交。你可在任意 IDE（如 Vim、VSCode）中审查变更。\n\n> 每次对话生成一个独立提交，支持按需回滚，无需手动 commit。","一位全栈开发者正在为公司内部的 Django 项目紧急修复一个生产环境的用户登录失败问题，时间紧迫，且代码库庞大、涉及多个模块（认证、会话、中间件），他需要快速定位并修复，同时避免引入新 bug。\n\n### 没有 codemcp 时\n- 需要反复在终端和 Claude Desktop 窗口间复制粘贴代码片段，容易出错或遗漏关键上下文。\n- 修改代码后必须手动运行 `python manage.py test`，再回到聊天窗口描述结果，流程碎片化，耗时超过 40 分钟。\n- 担心修改后忘记提交，或提交时混入无关改动，不得不频繁手动 git add \u002F commit，增加操作负担。\n- 无法直接让 AI 执行数据库迁移或清理缓存等命令，必须手动在终端输入，中断思考流。\n- 使用 VSCode 插件会强制绑定编辑器，而他习惯用 vim，无法无缝衔接工作流。\n\n### 使用 codemcp 后\n- 直接在 Claude Desktop 中说“修复用户登录时 SessionID 丢失的问题”，它自动读取项目代码、定位到 `auth\u002Fmiddleware.py` 并修改，无需手动复制。\n- 修改后自动运行测试套件，发现一个测试失败，它立即调整逻辑并重试，全程无需人工干预。\n- 所有改动自动提交到 Git 的独立分支，每一步修改都有独立提交记录，可随时 `git diff` 或 `git reset` 回滚。\n- 通过 `codemcp.toml` 预授权的命令，它能安全执行 `python manage.py migrate` 和 `redis-cli FLUSHALL`，无需切换终端。\n- 完全兼容 vim，修改完成后直接在熟悉的编辑器中审阅 diff，无需更换工具链。\n\ncodemcp 让 AI 从“代码翻译器”变成真正的编程协作者，把开发者从重复搬运和环境切换中解放出来，专注解决真正的问题。","https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fezyang_codemcp_803f314d.png","ezyang","Edward Z. Yang","https:\u002F\u002Foss.gittoolsai.com\u002Favatars\u002Fezyang_3d4f58b6.jpg",null,"New York, NY","ezyang@mit.edu","http:\u002F\u002Fezyang.com","https:\u002F\u002Fgithub.com\u002Fezyang",[84,88],{"name":85,"color":86,"percentage":87},"Python","#3572A5",99.7,{"name":89,"color":90,"percentage":91},"Shell","#89e051",0.3,1615,139,"2026-04-01T11:54:40","Apache-2.0","Linux, macOS, Windows","未说明",{"notes":99,"python":97,"dependencies":100},"需安装 Claude MCP 浏览器扩展并配置 SSE 服务端点；必须使用 Claude Pro 或 Claude Max 订阅；所有代码修改通过 Git 自动版本控制；需在项目根目录创建 codemcp.toml 配置文件以声明允许执行的命令；运行时需注意安全风险，禁止绑定到 0.0.0.0 以避免任意代码执行；推荐使用 Tailscale 安全远程访问。",[101,102],"uv","git",[15,45],"2026-03-27T02:49:30.150509","2026-04-06T05:16:58.238963",[107,112,117,122,127,132,137],{"id":108,"question_zh":109,"answer_zh":110,"source_url":111},9088,"如何解决 CodeMCP 在 Windows 上添加多余空格导致 Python 代码无效的问题？","该问题已在提交 acc00d933075cf2884d3b8b38272353c9bba582a 中修复。请更新到最新版本，确保使用包含该提交的代码，即可消除多余空格。","https:\u002F\u002Fgithub.com\u002Fezyang\u002Fcodemcp\u002Fissues\u002F143",{"id":113,"question_zh":114,"answer_zh":115,"source_url":116},9089,"CodeMCP 在空仓库（无提交）中报错，如何解决？","维护者已确认并修复了该问题，测试提交 33ed5c66f9e922dd0201b59b9cae692a9bcac380 已包含对空仓库的兼容处理。请更新到最新版本，问题将自动解决。","https:\u002F\u002Fgithub.com\u002Fezyang\u002Fcodemcp\u002Fissues\u002F158",{"id":118,"question_zh":119,"answer_zh":120,"source_url":121},9090,"如何避免 CodeMCP 遇到超时错误？","超时通常由 Anthropic 服务端延迟引起，而非 CodeMCP 本身。若出现长时间挂起，建议重启 Claude Desktop 客户端。若确认是 CodeMCP 服务端异常，可提供日志供进一步排查。","https:\u002F\u002Fgithub.com\u002Fezyang\u002Fcodemcp\u002Fissues\u002F144",{"id":123,"question_zh":124,"answer_zh":125,"source_url":126},9091,"如何创建一个支持 CodeMCP 的新项目？","当前 `RunCommand` 子工具需要 `chat_id` 参数，建议使用 `uvx` 命令启动 CodeMCP 服务以避免环境问题，例如：`uvx --from git+https:\u002F\u002Fgithub.com\u002Fezyang\u002Fcodemcp@prod codemcp serve`，确保服务正常运行后再进行项目初始化。","https:\u002F\u002Fgithub.com\u002Fezyang\u002Fcodemcp\u002Fissues\u002F162",{"id":128,"question_zh":129,"answer_zh":130,"source_url":131},9092,"如何让 CodeMCP 自动应用项目规范（如 Python\u002FFlask 最佳实践）而不需重复提示？","CodeMCP 主分支已支持 .cursor\u002Frules 文件，规则会在每次聊天和读取文件时自动注入。只需在项目根目录创建 .cursor\u002Frules 文件并写入规范，即可实现自动应用。","https:\u002F\u002Fgithub.com\u002Fezyang\u002Fcodemcp\u002Fissues\u002F3",{"id":133,"question_zh":134,"answer_zh":135,"source_url":136},9093,"为什么 `codemcp serve` 命令不启动服务？","该问题通常由 Python 环境冲突导致。建议使用 `uvx` 命令运行：`uvx --from git+https:\u002F\u002Fgithub.com\u002Fezyang\u002Fcodemcp@prod codemcp serve`，可避免全局 pip 安装的依赖冲突，确保服务正常启动。","https:\u002F\u002Fgithub.com\u002Fezyang\u002Fcodemcp\u002Fissues\u002F266",{"id":138,"question_zh":139,"answer_zh":140,"source_url":141},9094,"Claude 最新更新后不再显示 MCP 使用确认弹窗，是否影响 CodeMCP 功能？","无需担心，CodeMCP 设计为无需用户逐次确认工具使用。即使弹窗消失，功能仍正常运行。维护者已确认自动化流程不受影响，无需手动干预。","https:\u002F\u002Fgithub.com\u002Fezyang\u002Fcodemcp\u002Fissues\u002F283",[143],{"id":144,"version":145,"summary_zh":146,"released_at":147},106535,"0.6.0","This is a pretty big release with many days of commits since the last one, so it is going to be somewhat risky. The dependency footprint of the project also expanded substantially; I need to do some package splitting to get it back down but for now it's a little more bloated.\r\n\r\nFixed issues:\r\n* https:\u002F\u002Fgithub.com\u002Fezyang\u002Fcodemcp\u002Fissues\u002F253\r\n* https:\u002F\u002Fgithub.com\u002Fezyang\u002Fcodemcp\u002Fissues\u002F162\r\n* https:\u002F\u002Fgithub.com\u002Fezyang\u002Fcodemcp\u002Fissues\u002F198\r\n* Avoid shadowing glob module by renaming the glob_pattern.py file, thanks Lance Kindle\r\n\r\nFeatures:\r\n* Chmod tool to mark files executable (Sonnet really wants this tool to exist so I added it)\r\n* codemcp init - creates a new codemcp project with a codemcp.toml file. Also --python flag is supported to generate an opinionated Python scaffold (this is not as well tested and I think there's more scaffold I want to add). Thanks Lance Kindle for fixing some bugs in this implementation.\r\n* codemcp serve - serves codemcp as an SSE server, which is convenient for integration if you like running codemcp separately\r\n* You can now pass ~ inside your init codemcp prompt directory and it will work.\r\n* Trailing whitespace is always stripped when you write\u002Fedit files.\r\n* Some minor bugs in the prompt were fixed; unfortunately I don't have good evals to determine how much improvement this is worth. Thanks Tobias Mansfield-Williams for this fix.\r\n* We now properly report the current commit hash after all tool execution\r\n* I've done some mild testing with Goose + codemcp, and can confirm that you can use codemcp with Gemini 2.5 Pro this way, although I personally find it doesn't work as well as Sonnet 3.7\r\n\r\nDeveloper notes:\r\n* All of the e2e tests now no longer spawn a new process so you can properly pdb to debug issues now\r\n* The project is now fully typed with pyright!","2025-04-20T07:50:01"]