RAG Knowledge Base MCP Server

This is an MCP server I use to connect a RAG knowledge base to Claude Desktop, Claude Code, and other things.

Setup

uv sync --extra dev # installs pytest tests

Ingesting Documents

The Chroma backend does embedding and retrieval (you can use any sort of embedding/retrieval by implementing AbstractRagBackend)

Default configuration assumes the docs will be in knowledge-base in the project root.

Anything with that name is git-ignored, so one can create a symlink named knowledge-base pointing elsewhere.

The directory can also be configured by copying example.env to .env and...

RAG_KNOWLEDGE_DIR=./knowledge-base # <--- changing this to something else

What my directory looks like (more or less):

knowledge-base
├── companies
│   └── frobozz-co.md
│   └── acme.md
├── developers
│   └── wesley-hinkle.md
└── projects
|   ├── magic-api-gateway.md
|   ├── zork-legacy-cms.md
|   ├── torch-saas.md
|   ├── grue-detector.md
|   ├── zorkmid-sdk.md
|   ├── hello-footpad.md
|   ├── anvil.md
└── skills-mapping.md

Once the documents are in place:

uv run python ingest.py

or just

make embeddings

Client Configuration

Claude Desktop

Add to Claude Desktop config (~/Library/Application Support/Claude/claude_desktop_config.json on macOS):

{
  "mcpServers": {
    "rag-knowledge": {
      "command": "/absolute/path/to/uv",
      "args": [
        "run",
        "--directory",
        "/absolute/path/to/this/project",
        "python",
        "rag_knowledge_mcp.py"
      ],
      "env": {
        "LOG_LEVEL": "INFO"
      }
    }
  }
}

Note: Claude Desktop did not seem to have uv in its path so I used the absolute path returned from which uv

Claude Code

https://code.claude.com/docs/en/mcp

Installs with --scope local by default (claude in CWD)

claude mcp add rag-knowledge -- uv run --directory /absolute/path/to/this/project python rag_knowledge_mcp.py

Add --scope user for claude in any directory:

claude mcp add --scope user rag-knowledge -- uv run --directory /absolute/path/to/this/project python rag_knowledge_mcp.py

Before I managed to work out the syntax above I created a wrapper script:

claude mcp add --transport stdio rag-knowledge ~/.bin/rag-knowledge-mcp

The contents of wrapper-example.sh can be copied and modified

#!/usr/bin/env bash
set -euo pipefail

PROJECT_DIR="$HOME/path/to/this/project"
UV="$(which uv)"

echo "$PROJECT_DIR"
echo "$UV"

export LOG_LEVEL="${LOG_LEVEL:-INFO}"

cd "$PROJECT_DIR"
exec "$UV" run python rag_knowledge_mcp.py