zilliz

The Zilliz-MCP-Server provides a bridge between AI agents and Zilliz Cloud Milvus vector databases. It exposes two tool categories: Zilliz Control Plane Tools to manage Zilliz Cloud resources (projects, clusters, metrics) and Milvus Data Plane Tools to interact with data inside Milvus clusters (databases, collections, and vector operations). Agents can connect through standard MCP clients in two modes: a streamable HTTP service or via StdIO integration, allowing natural-language prompts to trigger provisioning, monitoring, and vector search workflows without writing code. Once connected, agents can issue commands like creating a free Milvus cluster, listing clusters, performing semantic search across a collection, inserting data, or querying cluster metrics, all through conversational prompts.

To use, configure your MCP client to connect to the Zilliz MCP Server (either via streamable-http or StdIO as shown in the examples). After connection, the client will present the available tools based on the selected transport, and you can start issuing natural-language requests such as: create a free Milvus cluster, run a semantic search on a collection, or retrieve cluster performance metrics. The server handles authentication, resource provisioning, and data-plane operations, returning results and connection details directly within the chat interface.

Prerequisites:

• Python 3.10 or higher
• uv (install via Homebrew on macOS: brew install uv, or install using the provided installer)
• A Zilliz Cloud account and API key

Installation steps:

1. Clone the repository: git clone https://github.com/zilliztech/zilliz-mcp-server.git cd zilliz-mcp-server

2. Install uv if not already installed (example for macOS): brew install uv

3. Prepare environment for streamable-http or StdIO usage:

   • Create an .env file (for streamable-http) and add your API key: ZILLIZ_API_KEY="your_api_key_here"
   • If using StdIO, ensure you have a ZILLIZ_CLOUD_TOKEN in your MCP client config as shown in the example.

4. Run the server in streamable-http mode (standalone HTTP service): uv run src/zilliz_mcp_server/server.py --transport streamable-http

5. Verify connectivity by configuring your MCP client to point to the server endpoint (e.g., http://localhost:8000/mcp) and confirm that available tools appear in your client tool list.

Environment variables and configuration tips:

• ZILLIZ_API_KEY is required for streamable-http mode to access Zilliz Cloud resources.
• ZILLIZ_CLOUD_TOKEN is used for StdIO-based integrations where the MCP client handles lifecycle and passes credentials to the server.
• When using StdIO, ensure the absolute path to the server project is correctly specified in your MCP client configuration.
• In streamable-http mode, ensure the server is reachable from your MCP client (default port is 8000) and that CORS and network policies allow access if needed.
• The Available Tools section in the README outlines the specific actions you can perform, such as list_projects, create_free_cluster, describe_cluster, insert_entities, search, and more. If a tool is missing from results, verify API keys and network connectivity.
• This MCP server is designed to work with Milvus vector databases through Zilliz Cloud and does not require direct Milvus administration; the server manages provisioning and access on your behalf.