# Knowledge Base MCP Server

An MCP server and command-line tool that makes a folder of Markdown notes searchable by meaning rather than only by keyword, splitting notes into chunks, embedding them with a locally or remotely configured model, and indexing the result in a local FAISS store.

Knowledge Base MCP Server reads a folder of Markdown notes, splits them into chunks, embeds each chunk with an embedding model of the user's choosing, such as a local Ollama model, OpenAI or Hugging Face, and stores the resulting vectors in a local FAISS index; if a local embedding model and a local language model are both used, notes never leave the machine, and only choosing a hosted provider sends anything over the network. Each subdirectory under a notes root is treated as its own knowledge base, or shelf, that can be searched individually or together.

Two interfaces sit on top of that index: an MCP server that any compatible client, such as Claude Desktop, Codex CLI, Cursor, Continue or Cline, can call to search notes, ask questions answered from them, and optionally write new notes back; and a kb command-line tool for the same corpus from a terminal or an agent's shell, which also exposes operational commands the MCP surface does not, such as indexing, diagnostics, evaluation and note curation. Retrieval quality can be tuned with optional reranking and gating stages, and the server can be exposed remotely over HTTP or server-sent events instead of only local stdio.

It requires Node.js 20 or later and is distributed under the Unlicense.

Website: https://github.com/jeanibarz/knowledge-base-mcp-server