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
With pytest tests:
uv sync --extra devConfiguration
Defaults live in the committed settings.yaml. Override them (all optional, merged in order, later wins):
settings.yaml committed defaults
~/.rag-knowledge-mcp/settings.yaml per-user override
settings.local.yaml per-machine override (gitignored)Under pytest, settings.test.yaml is merged over the defaults instead, so tests run against an isolated database.
Ingesting Documents
Embedding and retrieval are done with Chroma (other types of embedding/retrieval are possible by implementing AbstractRagBackend)
Default configuration assumes the KB docs will be in ./knowledge-base – which can be a symlink.
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.mdOnce the documents are linked up/in place:
uv run python ingest.pyor just
make embeddingsClient Configuration
Claude Desktop
Add to Claude Desktop config (~/Library/Application Support/Claude/claude_desktop_config.json on macOS):
{
"mcpServers": {
"rag-knowledge": {
"command": "uv",
"args": [
"run",
"--directory",
"/absolute/path/to/this/project",
"python",
"rag_knowledge_mcp.py"
],
"env": {
"LOG_LEVEL": "INFO"
}
}
}
}uv may need to be an absolute path, depending on how it's installed.
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.pyAdd --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
