arabold/docs-mcp-server

Fetches documentation from websites, GitHub, npm, PyPI, and local files, then indexes it with optional semantic vector search for AI-grounded queries—eliminating hallucinations by providing version-specific context. Implements MCP (Model Context Protocol) with SSE transport for long-running server mode or standalone CLI/Agent Skills, supporting 90+ file formats including PDFs, code repositories, and Jupyter notebooks. Runs entirely local with optional embedding models (OpenAI, Ollama, Gemini) for enhanced search quality.

1,127 stars and 4,033 monthly downloads. Actively maintained with 43 commits in the last 30 days. Available on npm.