Docs MCP Server by arabold: Setup for Claude, Cursor and Other Apps

• Searches documentation you have indexed

  Looks up passages in one indexed library and version, using keyword search and, if you set up an embedding model, search by meaning. Returns up to 5 matches by default.

  Try: “Search the react docs for useEffect cleanup.”

  Read · search_docs

• Reads a whole indexed page

  Returns the full text of one indexed page, in parts of up to 30,000 characters for long pages.

  Try: “Read the useEffect page from the react docs.”

  Read · read_page

• Lists the pages in an indexed library version

  Returns the indexed page paths, with an optional text filter and paging.

  Try: “List the pages in the react 18 docs that mention hooks.”

  Read · list_pages

• Lists the libraries you have indexed

  Shows which libraries are in your local documentation index.

  Try: “Which libraries have you indexed?”

  Read · list_libraries

• Finds the best matching indexed version

  Resolves a version or range such as 18.x to the closest version you have indexed.

  Try: “Which react version do you have for 18.x?”

  Read · find_version

• Fetches one web address as Markdown without indexing it

  Downloads a single page and returns it as Markdown, following redirects by default.

  Try: “Fetch https://react.dev/reference/react/useEffect as Markdown.”

  Network · fetch_url

• Indexes a documentation site, repo, package or folder

  Starts a background job that crawls a URL, GitHub repo, package or local folder and stores it in your local index, up to 1000 pages and 3 levels deep by default. A job ID comes back to follow progress.

  Try: “Index the react docs from https://react.dev/reference/react.”

  Write · scrape_docs

• Updates an indexed version, skipping unchanged pages

  Re-crawls an existing version and updates only the pages that changed.

  Try: “Refresh the react 18 docs.”

  Write · refresh_version

• Lists indexing jobs

  Shows queued, running and finished indexing jobs, filtered by status if you like.

  Try: “Show running indexing jobs.”

  Read · list_jobs

• Shows one indexing job

  Returns the status and progress of one job by its ID.

  Try: “How is job 3 going?”

  Read · get_job_info

• Cancels an indexing job

  Stops a queued or running job.

  Try: “Cancel the running react indexing job.”

  Write · cancel_job

• Removes an indexed library version

  Deletes a library version from your local index, so it can no longer be searched.

  Try: “Remove the react 17 docs from the index.”

  Delete · remove_docs

Watch out for

• It can index local folders and remove indexed docs

  The scrape tool takes local folders and archives as well as web pages, and the remove tool deletes an indexed version. Start it read-only if you only want to search. github.com

• Usage data goes to PostHog by default

  Telemetry is on in release builds. It sends tool names and some arguments such as library names, not full URLs or query text. Turn it off with the no-telemetry flag or DOCS_MCP_TELEMETRY=false. github.com

• Document text goes to your embedding provider

  If you turn on vector search, indexed text is sent to the provider you choose. A local one such as Ollama keeps it on your machine. github.com

• The web UI port is open on your machine

  Running the server in a terminal serves a web UI and the MCP endpoint on port 6280. The embedded stdio setup has no web UI. github.com

Security advisories