MDN Web Docs gets offline RAG-MCP server
A new open-source MCP server brings the entire MDN Web Docs library offline, using LanceDB for hybrid vector and full-text search. By running the embedding model locally via llama.cpp, it gives AI coding assistants zero-latency, private access to authoritative web development documentation.
Giving local LLMs direct access to MDN offline is a huge quality-of-life win for front-end workflows.
- –Hybrid search combining BM25 and 1024-d vectors ensures accurate retrieval of specific APIs and broad concepts alike
- –Local embedding generation via llama.cpp protects query privacy and eliminates API costs
- –The 50k+ row dataset is pre-ingested, meaning developers can start querying immediately without painful setup
- –Standard MCP integration makes it plug-and-play with modern AI IDEs and Claude Desktop
DISCOVERED
56d ago
2026-04-01
PUBLISHED
56d ago
2026-04-01
RELEVANCE
AUTHOR
dpswt