Local documentation sync and hybrid search (BM25 + vector) with URL citations.

• Node.js 20+ (ESM)
• npm

npm install

npm run build

The CLI is the primary entrypoint right now.

# Register a public documentation source
npm run cli -- add https://kubernetes.io/docs/ --name k8s --max-pages 500 --max-depth 5

# Sync the source (add --verbose for debug logs)
npm run cli -- sync <site_id> --verbose

# Search with citations
npm run cli -- search <site_id> "ServiceAccount what is it" --top-k 5

# Retrieve a full chunk by ID
npm run cli -- get-chunk <site_id> <chunk_id>

Notes:

• Only public (no-auth) sources are supported.
• Embeddings use fastembed's multilingual model (fast-multilingual-e5-large).
• Data is stored under ./data/ (sources metadata, hashes, LanceDB files).
• No-match searches return an empty results list.

src/mcp/server.ts exposes an in-process tool registry. A stdio MCP server is also available for the Inspector.

# Build the MCP server
npm run build

# Run MCP server over stdio (for Inspector)
npm run mcp

Inspector usage (opens a local UI and launches the server process):

npx @modelcontextprotocol/inspector node dist/mcp/stdio.js

• Configuration defaults live in src/config.ts.
• Build output is in dist/.