Milvus

Milvus is an MCP server that provides access to the Milvus vector database for Model Context Protocol clients like Claude Desktop and Cursor. It connects language models directly to local or remote Milvus database instances, allowing AI agents, software developers, and data engineers to manage collections and run multi-modal searches. Users can list existing collections, inspect schemas, create collections with customized vector dimensions, and insert document records through simple tool calls. The server enables sophisticated search capabilities, including dense vector similarity search, full-text lexical search, and combined hybrid searches with configurable filtering expressions and distance metrics such as COSINE, L2, or Inner Product. With support for standard input/output, Server-Sent Events, and streamable HTTP transports, this server exposes structured vector storage context directly into generative workflows, enabling retrieval-augmented generation and semantic memory management within developer workflows.

Category: AI Memory & Context

Tags: embeddings, milvus, vector-database, vector-search

Visit Milvus

How to install and configure Milvus

1. Ensure Python 3.10 or higher, a running Milvus instance, and uv are installed on your system. 2. Clone the repository using git clone https://github.com/zilliztech/mcp-server-milvus.git and navigate into the directory with cd mcp-server-milvus. 3. To run locally in stdio mode with Claude Desktop or Cursor, configure your MCP configuration file (claude_desktop_config.json or mcp.json): json { "mcpServers": { "milvus": { "command": "uv", "args": [ "--directory", "/path/to/mcp-server-milvus/src/mcp_server_milvus", "run", "server.py", "--milvus-uri", "http://localhost:19530" ] } } } 4. Alternatively, start the server over SSE using uv run src/mcp_server_milvus/server.py --sse --milvus-uri http://localhost:19530 --port 8000 and configure the client with the matching HTTP endpoint.

What you can do with Milvus

• Querying vector embeddings in Milvus collections using metric algorithms like COSINE or L2 to retrieve semantic context during conversational coding sessions. - Executing full-text and hybrid searches over stored text documents with custom filter expressions directly through natural language prompts in Claude Desktop. - Inspecting vector collection metadata and schema structures before dynamically inserting new embeddings and record attributes during retrieval pipeline debugging. - Creating new Milvus collections with defined vector dimensions, metric types, and field schemas without leaving the Cursor editor environment.

Key facts

• Open Source
• https://github.com/zilliztech/mcp-server-milvus
• AI Memory & Context, Databases & Data Stores
• embeddings, milvus, vector-database, vector-search

Part of MCP Servers

How do I install Milvus MCP server?

You can run the server directly using uv without manual installation. Clone the repository, ensure Python 3.10 or higher is installed, and point your client configuration to run server.py with your Milvus URI specified via the command line arguments.

What can Milvus MCP server do?

The server provides tools to list collections, inspect schemas, create collections, and insert data. It also allows clients to run vector similarity searches, keyword text searches, and hybrid searches using customizable distance metrics, limits, and filter expressions.

Which MCP clients work with Milvus?

It works with any client supporting the Model Context Protocol, including Anthropic Claude Desktop, the Cursor code editor, and custom client implementations. It supports standard input/output, SSE, and streamable HTTP communication modes.

Is Milvus MCP server open source?

Yes, the Milvus MCP server is open source. You can inspect the source code, review updates, and contribute to the repository hosted by Zilliz on GitHub.