mcp-tool-server-milvus

What is mcp-tool-server-milvus?

mcp-tool-server-milvus is an MCP server that provides tool functions to insert data into and query data from the Milvus vector database. It is designed for developers who want to integrate AI‑powered vector search into their applications.

How to use mcp-tool-server-milvus?

Run docker compose up -d to start the server on http://localhost:3000. Then interact using npx @wong2/mcp-cli --url http://localhost:3000/mcp or the simple streamable HTTP client from the MCP TypeScript SDK. By default, the server creates a database named mcp_db and a collection named full_text_search. These can be overridden with the environment variables MILVUS_DB and MILVUS_COLLECTION.

Key features of mcp-tool-server-milvus

• Provides three tools: insert, insert-multiple, and search.
• Automatically creates a default database and collection.
• Configurable database and collection names via environment variables.
• Runs as a Docker container with a single command.
• Supports interaction via MCP-Cli or a streamable HTTP client.
• Returns the number of inserted records from insert-multiple.

Use cases of mcp-tool-server-milvus

• Insert individual text documents into a Milvus vector collection for later retrieval.
• Batch-insert multiple texts using the insert-multiple tool.
• Perform semantic similarity search by submitting a query text to the search tool.
• Prototype and test MCP‑based vector search workflows locally.

FAQ from mcp-tool-server-milvus

Does mcp-tool-server-milvus require a running Milvus instance?

Yes, the server runs with Milvus as its backend. The README uses Docker Compose to start both the MCP server and Milvus together, so a separate Milvus installation is not needed when using the provided setup.

How do I configure the database and collection names?

Set the environment variables MILVUS_DB (default: mcp_db) and MILVUS_COLLECTION (default: full_text_search) before starting the server. The database and collection are created automatically if they do not exist.

What tools does the server expose?

Three tools are available: insert (insert a single text string), insert-multiple (insert an array of text strings), and search (search for matching texts). All operate on the configured database and collection.

Can I interact with the server without the MCP-Cli tool?

Yes. The README describes using the simpleStreamableHttp example from the @modelcontextprotocol/sdk package. Clone the SDK repository, run the example client, and use commands like list-tools and call-tool to interact.

Is authentication required or supported?

The README does not mention any authentication or authorization mechanisms. It assumes a local, unsecured setup.