MCP Server for Milvus

What is Milvus?

MCP Server for Milvus provides access to the Milvus vector database through the Model Context Protocol. It enables LLM applications like Claude Desktop and Cursor to search, manage, and query vector data using natural language or programmed instructions.

How to use Milvus?

Run the server directly with uv via uv run src/mcp_server_milvus/server.py --milvus-uri http://localhost:19530. Configure it as an MCP server in Claude Desktop or Cursor by specifying the command and arguments. Alternatively, set environment variables MILVUS_URI, MILVUS_TOKEN, and MILVUS_DB.

Key features of Milvus

• Full-text search on collections with drop ratio control
• Vector similarity search (COSINE, L2, IP metrics)
• Hybrid search combining vector and attribute filtering
• Collection management: create, load, release, list, info
• Data operations: insert, bulk insert, upsert, delete
• Index creation and index info retrieval

Use cases of Milvus

• Searching documents in a collection using natural language via Claude Desktop
• Performing vector similarity searches for recommendation or retrieval systems
• Managing Milvus collections and schemas directly from Cursor Composer
• Inserting and updating data in Milvus through an AI assistant

FAQ from Milvus

What are the prerequisites for using this server?

Python 3.10 or higher, a running Milvus instance, and uv installed.

How do I connect to a remote Milvus instance?

Pass the URI via the --milvus-uri argument or set the MILVUS_URI environment variable.

What if the MCP tools do not appear in Claude Desktop or Cursor?

Restart the application after configuring the MCP server; also try pressing the refresh button in Cursor’s MCP settings.

How do I handle authentication with Milvus?

Set the MILVUS_TOKEN environment variable with your authentication token.

What transport does the server use?

It uses stdio transport, as shown in the Cursor configuration with “Type: stdio”.