LanceDB

What is LanceDB?

LanceDB is a Node.js MCP server that connects to a LanceDB database and performs vector similarity search using Ollama’s embedding model. It is intended for developers who want to integrate vector search into AI workflows or build retrieval-augmented generation (RAG) applications.

How to use LanceDB?

Install dependencies with pnpm install, then run the vector search test script using pnpm test-vector-search or node test-vector-search.js. To use as an MCP service with Claude Desktop, add the provided JSON configuration to your MCP settings, replacing the paths with your actual installation and storage directories.

Key features of LanceDB

• Connects to a LanceDB database for vector storage
• Custom embedding function using Ollama’s nomic-embed-text model
• Performs vector similarity search against stored documents
• Processes and displays search results with similarity scores
• Integrates with Claude Desktop as an MCP service
• Uses 768‑dimensional embeddings

Use cases of LanceDB

• Searching a document collection for semantically similar passages
• Building a RAG pipeline with local embeddings and vector storage
• Demonstrating vector search integration with Ollama and LanceDB

FAQ from LanceDB

What are the prerequisites to run LanceDB?

Node.js v14 or later, Ollama running locally with the nomic-embed-text model, and a LanceDB storage location with read/write permissions.

How do I configure LanceDB for Claude Desktop?

Add the following to your MCP configuration JSON, replacing the paths with your actual locations:

{
  "mcpServers": {
    "lanceDB": {
      "command": "node",
      "args": ["/path/to/lancedb-node/dist/index.js", "--db-path", "/path/to/your/lancedb/storage"]
    }
  }
}

What embedding model does LanceDB use, and what are its dimensions?

It uses Ollama’s nomic-embed-text model, which produces embeddings with 768 dimensions.

Where is the vector data stored?

Data is stored in the LanceDB directory specified by the --db-path argument in the MCP configuration (or the default path used in the test script).

What dependencies does LanceDB require?

The project depends on @lancedb/lancedb (LanceDB client), apache-arrow, and node-fetch for making API calls to Ollama.