mcp -milvus

This MCP server provides access to Milvus vector database functionality via the MCP interface. It exposes a set of tools for performing text and vector searches, hybrid queries, collection management, and data operations against a Milvus instance. You can connect compatible MCP clients (like Claude Desktop, Cursor, or other MCP-enabled apps) and issue standardized tool calls such as milvus-text-search, milvus-vector-search, milvus-hybrid-search, milvus-multi-vector-search, milvus-query, milvus-count, milvus-list-collections, milvus-create-collection, milvus-insert-data, and more. The server is designed to run behind uv and expects a Milvus endpoint to be reachable at the specified URI.

To use with clients, configure the MCP server entry under the client’s MCP settings with the appropriate command and arguments, including the Milvus URI. Example usage will typically provide the uv command to run the server script and pass the Milvus endpoint as a parameter, allowing the client to discover and invoke Milvus tools through the MCP protocol.

Prerequisites

• Python 3.10 or higher
• A running Milvus instance (local or remote)
• uv installed (recommended for running the server)

Installation steps

1. Install uv (if not already installed):
   • pip install uv
2. Ensure Milvus is running and accessible at the desired URI (e.g., http://localhost:19530).
3. Clone the repository and install any additional dependencies if needed:
   • git clone https://github.com/stephen37/mcp-server-milvus.git
   • cd mcp-server-milvus
   • pip install -r requirements.txt # if a requirements file exists
4. Run the MCP server with uv using the Milvus endpoint:
   • uv run src/mcp_server_milvus/server.py --milvus-uri http://localhost:19530

Notes

• Adjust the Milvus URI as appropriate for your environment (remote Milvus, dockerized Milvus, etc.).
• If you use Docker or other environments, adapt the command to your deployment method while preserving the same server entry point andMilvus URI.

Tips and caveats:

• The Milvus URI is central; ensure the Milvus instance is reachable from the MCP server process. Network or firewall issues can block access.
• When using Cursor or Claude Desktop, ensure the MCP server configuration points to the correct uv command and Milvus URI. Use 127.0.0.1 instead of localhost where DNS resolution could cause issues.
• Validate tool availability in your Milvus instance by listing collections and performing a basic vector search to confirm connectivity.
• If you encounter issues, check uv server logs for connection errors to Milvus, and verify that the Milvus port is open and accepting connections.
• The available tools cover common Milvus operations: text search, vector search, hybrid search, multi-vector search, arbitrary queries, and collection management. Explore and tailor the parameters (collection_name, vector, query_text, filter_expr, etc.) to your data model.