Lancedb

🗄️ LanceDB MCP Server for LLMs

A server based on the Model Context Protocol (MCP) that enables LLMs to directly interact with locally stored documents, implemented in LanceDB through proxy RAG and hybrid search. You can ask LLMs questions about the entire dataset or specific documents.

🚀 Quick Start

To get started, create a local directory to store the index and add the following configuration to your Claude Desktop configuration file:

MacOS: ~/Library/Application\ 支持/Claude/claude_desktop_config.json
Windows: %APPDATA%/Claude/claude_desktop_config.json

{
  "mcpServers": {
    "lancedb": {
      "command": "npx",
      "args": [
        "lance-mcp",
        "Path to the local index directory"
      ]
    }
  }
}

Prerequisites

• Node.js 18+
• npx
• MCP client (e.g., Claude Desktop App)
• Summary and embedding models installed (see config.ts, Ollama models are used by default)
  • ollama pull snowflake-arctic-embed2
  • ollama pull llama3.1:8b

Example

Local Development Mode:

{
  "mcpServers": {
    "lancedb": {
      "command": "node",
      "args": [
        "Path to the generated index.js file"
      ]
    }
  }
}

You need to run npm run build to generate index.js.

Use the Inspector tool to check the status of the MCP server:

inspect http://localhost:3000

Seed Data

Run the seed script to create two tables:

1. Document summaries.
2. Chunked vector indexes.

When running the command, specify the database path and the location of the document folder:

node seeds/index.js --db <Database path> --docs <Document folder path>

If you need to recreate the index, you can use the --overwrite option.

Example Prompts

• Query documents in the catalog: “What documents do we have in the catalog?”
• Query based on content: “Why is the US healthcare system so broken?”

Tools

• Catalog Tools:
  • Search for summaries.
• Chunks Tools:
  • Search for specific document chunks.

✨ Features

• 🔍 Serverless vector indexing and document summary catalog based on LanceDB.
• 📊 Efficient use of LLM tokens. LLMs find the required information by themselves when needed.
• 📈 Security. The index is stored locally, so no data is transferred to the cloud when using local LLMs.

📄 License

This project uses the MIT License, which allows for free use and modification.