Access LLM Documentation Locally or Remotely

DocsMCP is an MCP server that gives an LLM access to documentation, whether that documentation lives in local files or at remote URLs. Sources are configured up front, with a name and a URL or path, and the server exposes them to any connected MCP client so the LLM can query documentation directly instead of relying on what it already knows.

Use it when an AI assistant needs to reference specific, current documentation - a protocol spec, a library's docs, or an internal reference file - rather than answer from training data that may be outdated or incomplete. It suits both single-source setups, one named documentation source configured at startup, and ad hoc lookups, since getDocumentation can fetch and parse any given URL or file path directly rather than only sources declared at launch. It is a documentation retrieval layer, not a search engine or a code execution tool - it fetches and parses content, it does not index or rank multiple sources for relevance itself.

getDocumentationSources lists all documentation sources that have been configured for the running server. getDocumentation fetches and parses documentation from a given URL or local file path, taking a single url parameter that can point to either.

For Cursor, add a .cursor/mcp.json file:

{
  "mcpServers": {
    "docs-mcp": {
      "command": "npx",
      "args": [
        "-y",
        "docsmcp",
        "--source=Model Context Protocol (MCP)|https://modelcontextprotocol.io/llms-full.txt"
      ]
    }
  }
}

Wrap the source string in quotes whenever the source name contains spaces. For VS Code, the equivalent configuration goes in .vscode/mcp.json under the stdio type, using the same npx docsmcp command and --source argument.

Developers who want an LLM in their editor grounded in specific, current documentation, such as a protocol, library, or internal reference, instead of guessing from training data. The project is licensed under MIT.