DevDocs MCP

What madhan-g-p/DevDocs-MCP MCP server does

madhan-g-p/DevDocs-MCP MCP server gives MCP-compatible coding assistants access to locally cached documentation sourced from DevDocs.io. Its purpose is to provide a stable documentation reference for a project rather than generate code, act as an AI agent, scrape arbitrary websites, or replace the DevDocs user interface.

The server is intended for projects where dependency versions matter. A project can provide dependencies such as React 18.2.0 and TypeScript 5.0.0, allowing documentation context to be associated with the versions used by that project. This helps avoid mixing APIs from different releases.

Documentation is obtained through DevDocs' supported offline-data mechanism. The project description and README state that it uses structured DevDocs datasets rather than web scraping.

How it works

madhan-g-p/DevDocs-MCP MCP server runs as a local Node.js application written in NestJS and TypeScript. Its data flow separates metadata from document content:

• SQLite, accessed through sql.js, stores the metadata database.
• JSON files on local disk hold cached documentation.
• Documentation can be fetched lazily from the DevDocs.io API when it has not yet been ingested.
• Ranked fuzzy search helps locate relevant entries such as “intersection observer.”

After ingestion, the cached documentation can be used without an internet connection. The README describes the returned content as structured for language-model use. The architecture also shows agent requests being handled through MCP search and explain operations, while ingest is the documented tool for downloading documentation for a project stack.

Setup and configuration

The documented source installation uses Node.js 18 or newer, with testing noted for Node.js 18.x and 20.x. Clone the repository, run pnpm install, build it with pnpm build, and start the production process with pnpm start:prod. The project asks contributors to use pnpm for development.

Configuration is read from a .env file. Copy .env.example to .env, then set storage locations as needed. DEVDOCS_DATA_PATH controls where downloaded documentation JSON files are kept and defaults to ./data. MCP_DB_PATH controls the SQLite metadata database path and defaults to mcp.db. LOG_LEVEL controls logging verbosity and defaults to info.

Docker is also documented. The published image is madhandock1/devdocs-mcp:latest, and the recommended persistent mount is /app/data. That directory contains both mcp.db and the cached documentation under docs/.

For STDIO clients such as Claude Desktop and Cline, the example starts node dist/main.js from the built project directory. For HTTP SSE use, start the same entry point with --port 3000; the documented SSE path is /mcp/sse.

Tools and capabilities

The documented ingest tool downloads documentation for dependency versions supplied by the agent. The server also supports documentation lookup through ranked fuzzy search and exposes search/explain behavior in its MCP interaction model. Its main capabilities are local caching, version awareness, offline access after ingestion, structured output, and a Node-only runtime without native Python or C++ build steps.

Limitations and notes

madhan-g-p/DevDocs-MCP MCP server is explicitly marked as being under active development. The README notes that caching, automatic updates, on-demand updates, and query behavior may change or receive further implementation. APIs and design decisions may therefore evolve.

The initial ingestion step requires access to the DevDocs data source. Offline operation applies after the relevant documentation has been downloaded. The README also notes that the fetch URL is defined in src/config/constants.ts; if DevDocs changes its supported path, that configuration may need an update. The project is local software, not a hosted SaaS endpoint, although it can be exposed over HTTP SSE when started with a port.