Milvus MCP Server – Vector Database Operations
The official MCP Server for Milvus lets AI applications interact with a running Milvus vector database using natural-language workflows. Use it to manage collections, insert and retrieve data, run vector searches, inspect schemas, and query database state from an MCP-compatible client.
Overview
The MCP Server for Milvus connects compatible AI agents to a running Milvus
vector database. It lets agents perform common vector-database operations
through MCP tools, making Milvus accessible from clients such as Claude Desktop,
Cursor, or custom MCP-compatible applications without writing bespoke database
integration code for each assistant.
What the MCP server enables
The server is designed for direct interaction with Milvus collections and data.
Depending on the configured Milvus connection and credentials, an AI agent can:
- Connect to a local or remote Milvus instance.
- List and inspect collections.
- Create and manage collections.
- Insert data into collections.
- Retrieve stored entities.
- Run vector similarity searches.
- Execute structured queries over collection data.
- Work with database and collection configuration supplied through CLI options
or environment variables. - Use natural-language prompts to explore collection schemas and search results.
When to use it
Use Milvus MCP when an AI agent needs live access to vector-search data or must
help operate a Milvus-backed application. Practical examples include inspecting
a RAG collection, testing similarity search behavior, loading sample embedding
records, reviewing collection structure, querying stored metadata, debugging
retrieval quality, or letting a coding assistant validate Milvus operations
against a real development database.
Connection and authentication
The recommended setup runs the server locally with uv. From a cloned
zilliztech/mcp-server-milvus checkout, the documented default command is
uv run src/mcp_server_milvus/server.py --milvus-uri http://localhost:19530.
Stdio is the default transport and is suited to single-client desktop or IDE
integrations.
SSE mode can be enabled with --sse and a port, such as
uv run src/mcp_server_milvus/server.py --sse --milvus-uri http://localhost:19530 --port 8000.
If the target Milvus instance requires token authentication, provide the token
through the documented environment or command-line configuration instead of
hardcoding it in client files.
Key considerations
The MCP server operates against a real Milvus instance, so tool calls can change
data and collections when write-capable tools are used. Use development
databases for experimentation, protect Milvus tokens, and require review before
insert, collection-management, or destructive operations. The server requires a
reachable Milvus instance and does not replace normal Milvus access controls,
backup planning, schema design, or production change-management processes. For
HTTP-style access, restrict SSE endpoints to trusted networks and clients.
Supported Transports
stdio
Command: uv
Args:
runsrc/mcp_server_milvus/server.py--milvus-uri${MILVUS_URI}
sse
URL: http://localhost:8000/sse
Frequently Asked Questions
- When should an AI agent use the Milvus MCP server?
- Use it when a workflow needs live Milvus vector-database context or approved operations, such as inspecting collections, inserting records, retrieving entities, running vector search, querying metadata, or debugging retrieval behavior in a RAG system.
- What does the Milvus MCP server add to an AI agent's capabilities?
- It gives the agent structured access to a running Milvus instance, allowing it to perform database and vector-search operations instead of relying only on static model knowledge or manually copied query results.
- What can an AI agent access or manage through Milvus MCP?
- Depending on server configuration and Milvus permissions, the agent can inspect and manage collections, insert and retrieve data, run vector similarity searches, execute queries, and review collection or database state exposed by the official server tools.
- How is authentication configured for the Milvus MCP server?
- Configure the server with the Milvus URI and, when the target instance requires authentication, provide MILVUS_TOKEN or the documented command-line token option to the server process. Store tokens securely and never commit them to source control.
- Which transport should be used for the Milvus MCP server?
- Use stdio for local desktop and IDE clients such as Claude Desktop or Cursor. Use SSE only when an HTTP-style or multi-client setup is needed, and keep the SSE endpoint restricted to trusted clients and networks.