Skip to content
Kitbay
MCP server

knowledge-rag

Runs locallyMITUpdated on Nov 15, 2026306GitHub starsCan change dataNo API key needed

A local MCP server that indexes documents and code on your own machine so an assistant can search them.

What it does

A local MCP server that indexes documents and code on your own machine so an assistant can search them. It combines keyword and vector search and reranks results, and it reads about 33 file types by default, including Markdown, PDF text, Word, Excel, PowerPoint, notebooks and common source files. It only retrieves passages; the client's own model writes the answer. Embeddings run locally, so no API key is needed after the first model download. Tools cover searching, reading and listing documents, adding or updating them (also from a URL) and reindexing. It runs over stdio, or over HTTP for several clients, and needs Python 3.11 or later.

Capabilities

  • Index local documents and code for search
  • Search with combined keyword and vector matching
  • Add, update or remove documents in the index
  • Add a document from a URL