agentdocs

io.github.titov-d由 GitHub 账号 titov-d 发布v0.1.3更新于 Oct 24, 2026

Fresh, source-verified docs for building MCP servers and Claude agents, for AI coding agents.

已验证STDIO仅桌面知识与记忆开发者工具

概览

AI 生成的概览

本地 MCP 服务器,为编程智能体提供经过来源核验的 MCP 服务器与 Claude 智能体开发文档语料,可检索而非靠猜测。

可以做什么
  • 列出全部精选文档及其主题、来源网址与核验状态
  • 在语料中检索,返回按相关度排序并标注来源的片段,可按主题缩小范围
  • 按 id 读取单篇文档完整的已核验 Markdown 内容
适用场景
适合在助手针对快速变化的 API(如 Model Context Protocol 或 Claude Agent SDK)编写代码时使用,此时训练数据落后于规范,容易产生错误导入或过时写法。它的用途是让助手先查询再作答,而不是凭记忆回答。
运行要求
以 npm 包 mcp-agentdocs 通过 stdio 在本地运行,需要 Node.js 18 或更高版本,通常用 npx 启动。无需账号、API 密钥或托管服务。
安装前请注意
v1 语料仅覆盖 MCP 规范以及构建 MCP 服务器和 Claude 智能体,并会逐步扩充,因此不代表覆盖其他主题。核验标记只反映某一时点的人工审核,回答仍应回到所引用的上游来源核对。

安装

在 SourceWeft 中

  1. 打开 控制台中的 agentdocs,将其添加到工作区。
  2. 为需要使用其工具的对话启用该服务。

Desktop only,通过 STDIO。 STDIO 服务会启动本地进程,因此需要 SourceWeft 桌面宿主。

其他 MCP 客户端

参照 仓库 中的启动说明。

README

agentdocs

Fresh, source-verified docs for building MCP servers and Claude agents — served straight to your AI coding agent over MCP.

[npm] [license: MIT] [Model Context Protocol]

Coding agents hallucinate on fast-moving APIs like the Model Context Protocol and the Claude Agent SDK: training data lags the spec, so you get plausible-but-wrong imports, deprecated patterns, and invented options. agentdocs is a local MCP server that gives your agent a curated, source-verified corpus to search instead of guessing.

  • 🔎 Three tools: list_topics, search_docs, get_doc
  • 📎 Every answer is attributed to its upstream source URL
  • 🏠 Runs locally over stdio — no account, no hosting, no data leaves your machine
  • ✅ Each doc carries a verified flag set only after human review against the source

Install & connect

Requires Node.js 18+.

Claude Code

bash
claude mcp add agentdocs -- npx -y mcp-agentdocs@latest

Cursor / other MCP clients (mcp.json)

json
{  "mcpServers": {    "agentdocs": {      "command": "npx",      "args": ["-y", "mcp-agentdocs@latest"]    }  }}

Then ask your agent things like "check agentdocs for how MCP stdio handles logging" or "search agentdocs for the Streamable HTTP session header" — it will call the tools and answer from the verified corpus.

Tools

ToolUse it to
list_topicsSee every curated doc, its topic, source URL, and verified status
search_docs(query, topic?)Get ranked, source-attributed snippets for a question
get_doc(id)Read the full verified markdown of one document

What's covered (v1)

The Model Context Protocol (spec 2025-11-25) and building MCP servers / Claude agents. Corpus grows over time; run list_topics for the current set.

Freshness

Every document records the date it was last verified against its primary source. Sources are diffed on a regular cadence; when one drifts, the affected doc is re-verified and a new version is published. Using mcp-agentdocs@latest (as above) keeps you on the current corpus. See CHANGELOG.md for what changed and when.

Development

bash
npm installnpm run build      # tsc -> dist/npm test           # vitest: retrieval, manifest integrity, unicode safetynpm run ingest     # refresh sources into corpus/_staging/ for human review

Curation is the point: npm run ingest fetches upstream sources for review but never silently overwrites a curated doc — a human diffs, promotes, and sets verified: true in corpus/manifest.json only after checking the content against its source. Run npm run check-freshness to flag docs that have drifted or are due for re-verification.

License

MIT

来源:README.md,提交 7915d8c

工具

0
SourceWeft 尚未在沙箱中启动过该软件包,因此其工具未知。

版本历史

1
  1. v0.1.3最新Sep 16, 2026

评分

已满足适用检查项的 50%

该评分衡量此收录条目的完整程度,依据注册中心条目、端点与公开来源计算得出。它不是对服务器本身的评价,也不表示服务器安全或当前可用:可连通性是带日期的检查结果,评分也不包含安装量或用户评价。

  • 近 30 天内可连通25 分

    请保持端点能够响应 MCP 握手;SourceWeft 每 30 天探测一次。

  • 已声明认证方式10 分

    请在注册中心条目中声明凭据请求头,或保持端点能够响应探测,以便判定其认证方式。

  • 已列出工具15 分

    请让服务器在无凭据时响应 tools/list,或发布可在沙箱中无凭据启动的软件包。

  • 描述必需15 分
  • 图标10 分
  • 文档15 分
  • 许可证10 分