lancedb

This LanceDB MCP server provides a serverless interface that uses LanceDB to store and retrieve data via three tools: Ingest docs, Retrieve docs, and Get table details. The server is designed as a reference implementation for building MCP apps with LanceDB, enabling you to ingest documents into LanceDB, query stored documents, and inspect table metadata. The tooling allows you to index knowledge into LanceDB for efficient retrieval and to surface table information for debugging and validation. Use the Ingest docs tool to embed and store docs, the Retrieve docs tool to run queries against your indexed data, and the Get table details tool to inspect table schemas and stats for your LanceDB storage.

Prerequisites:

• Python 3.8+ and a Python environment
• Access to LanceDB (via Python package) or a local LanceDB installation
• uv (uvx) installed and available in your PATH
• The MCP server script at /path/to/your/mcp/lancedb_mcp.py (or your chosen path)

1. Install Python dependencies (example using pip):

pip install lancedb uvx

2. Ensure LanceDB directory is prepared and accessible:

# Example (adjust to your environment):
mkdir -p /Path/to/your/lancedb_mcp

3. Create or update your Claude MCP config with the following (as shown in the README):

{
  "mcpServers": {
    "lancedb": {
      "command": "uv",
      "args": [
        "--directory",
        "/Path/to/your/lancedb_mcp",
        "run",
        "/path/to/your/mcp/lancedb_mcp.py"
      ]
    }
  }
}

4. Run the MCP server via your preferred method (naming depends on your setup):

# If using the provided config, uv will start the MCP server as configured.

5. Verify the server is listening and accessible by issuing a test query (via your MCP client).

Tips and notes:

• Update the paths in the mcp_config to reflect your actual LanceDB and MCP script locations.
• If you encounter permission or directory access issues, ensure the user running the MCP process has read/write access to /Path/to/your/lancedb_mcp and the Python script.
• LanceDB stores data locally; consider backing up the LanceDB directory regularly for resilience.
• The three tools are designed to be simple references; you can extend or replace the underlying LanceDB usage with more advanced indexing or query patterns as needed.
• If you need to adjust environment variables (e.g., Python path, LanceDB settings), you can add an env section where supported by your deployment environment.