Context Atlas

io.github.levin-staudt由 GitHub 账号 levin-staudt 发布v1.0.0更新于 Oct 24, 2026

Find llms.txt, docs, MCP servers, skills and CLIs for a tech stack, for coding agents.

发布方自己的服务:Red Atlas Api

已验证Streamable HTTP可网页运行知识与记忆开发者工具

概览

AI 生成的概览

一个远程 MCP 服务器,收录厂商为编码智能体发布的 llms.txt、文档、MCP 服务器、技能、规则和 CLI,可按工具或技术栈检索。

可以做什么
按工具名称、角色、区域或检测信号(文件、依赖、环境变量)搜索目录,每个词按前缀匹配。按 id 获取单个工具的全部来源、检测信号和角色。列出目录中所有工具的 id、名称、角色和区域,便于按角色寻找候选。通过查看各工具的 llms.txt 或规则来源进行比较。
适用场景
当编码助手需要为现有技术栈找到合适的面向智能体的来源,或想按角色比较候选工具、再为新项目推荐技术栈时适用。
运行要求
远程 Streamable HTTP 端点;未声明任何软件包、账户、API 密钥或环境变量。需要能访问该服务的网络。
安装前请注意
目录仅链接到厂商来源,内容归各厂商所有,应到原始来源核实。全文来源有数 MB,服务器建议先获取某个主题的 llms.txt 或规则来源。该服务器为只读且无状态。

安装

在 SourceWeft 中

  1. 打开 控制台中的 Context Atlas,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。

Web executable,通过 Streamable HTTP。 远程服务在工作区中配置后即可从网页运行时运行。

交给智能体安装

把这段提示词粘贴到 Claude Code、Codex、Cursor 等智能体,或 SourceWeft 对话中。智能体会读取该服务器的安装指南,先向你展示发布者、端点和登录方式,经你同意后再把它添加到你的客户端。密钥不会写进任何命令。

阅读 https://sourceweft.com/mcp/io.github.levin-staudt%2Fcontextatlas/install.md,并按其中说明把这个 MCP 服务器接入我的客户端。修改任何配置之前,先告诉我它的发布者、端点和登录方式,等我确认。不要把 API 密钥写进命令或我会提交的文件;问我密钥应从哪里获取。

查看智能体所依据的安装指南

其他 MCP 客户端

把它添加到你客户端的 mcpServers 配置中。

{
  "mcpServers": {
    "context-atlas": {
      "type": "http",
      "url": "https://contextatlas.dev/mcp"
    }
  }
}

README

Context Atlas

Live: https://contextatlas.dev (the old https://contextatlas.levinstaudt.workers.dev redirects there with a 301)

Chart a tech stack, or find the right sources for the one you have. Context Atlas collects the llms.txt indexes, docs, MCP servers, agent skills, rule files and CLIs that vendors publish for coding agents, grouped into cloud providers, frameworks, standards and CLIs.

The site only links to the sources; every source belongs to its vendor.

  • / lists every source, filterable by search and by kind (the map legend).
  • /agent has a copy button for a one-sentence prompt, either for an existing project (detect the stack, then follow the matching sources) or for a new one (propose a stack), and the command to add the MCP server.
  • /llms.txt holds the agent instructions and the full catalog; the prompt only links to it. /atlas.json has the same data as JSON.
  • /llms/<topic-id>.txt holds a single topic, for agents that need only one tool's sources.
  • /mcp is a stateless MCP server (Streamable HTTP) with the tools search, get_topic and list_topics: claude mcp add --transport http contextatlas https://contextatlas.dev/mcp.
  • /api/topics, /api/topics/<id> and /api/search?q=<words>&kind=<kind>&limit=<n> are the JSON API for the CLI (#13).

Add a source

All content lives in packages/core/src/catalog.ts. Add a Source to an existing topic, or a new Topic with a role and signals (how an agent detects it in a repo: file(...), dep(...), env(...) or hint(...)). Bump updatedAt. Then run bun run check; the catalog tests catch duplicate ids, bad URLs and malformed signals. The add-source skill has the details.

Everything the site serves is read from SQLite, which bun run seed (and every deploy) rebuilds from that file.

Develop

Bun workspaces, Hono in front for the agent files, the API and MCP, React Router (framework mode, server-rendered, Vite) for the pages, SQLite through Drizzle with FTS5 search, the MCP TypeScript SDK and Tailwind CSS.

bash
bun installbun dev          # http://localhost:3000, hot reload for the pages, restart on server changesbun run check    # lint, typecheck, tests: run before every commitbun run build    # dist/: single binary (server + pages) + public/ + drizzle/bun run deploy   # deploy main to the VPS through GitHub Actions
PathWhat
packages/coreCatalog, database schema and migrations, seed, queries, search, llms.txt text. Transport-free.
apps/serverHono app: agent files, JSON API, MCP, static files; hands every other path to apps/web. src/main.ts is the binary (seed, serve), src/dev.ts the dev server.
apps/webThe pages for people: React Router routes, components and styles. Reads the catalog from the server's snapshot, no data logic. See docs/web-frontend.md.
apps/redirectCloudflare Worker that 301s the old workers.dev URL to the new domain.
deploy/systemd unit and VPS scripts.
scripts/Cloudflare setup, VPS access, IndexNow ping and the share image, all from the harness.
server.json, glama.jsonThe remote MCP server's entries for the official MCP Registry and Glama.
docs/deploy.mdInfrastructure and runbook.

来源:README.md,提交 97845e8

工具

3

get_topic

Get a topic

All sources (llms.txt, docs, MCP servers, skills, rules), detection signals and the role of one tool, by id such as `vercel`, `nextjs` or `drizzle`.

输入参数id必填

list_topics

List topics

Every tool in the catalog with its id, name, role and region, without sources. Use it to find candidates per role, then get_topic and their llms.txt sources to compare them.

输入参数region

版本历史

1
  1. v1.0.0最新Oct 24, 2026

评分

已满足适用检查项的 90%

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 天内可连通必需25 分
  • 已声明认证方式必需10 分
  • 已列出工具必需15 分
  • 描述必需15 分
  • 图标10 分
  • 文档15 分
  • 许可证10 分

    请在仓库或注册中心条目中声明许可证。

其他 Atlas 集成