mcp-tool-server-milvus

A MCP server which provides tools function to insert into and query data from the Milvus vector database.

By default, the server creates a database and a collection. The insert and search tools operate upon the default database and collection. You can use the following env vars to configure the database and collection:

MILVUS_DB: The name of the Milvus database. Default is mcp_db. The db will be created if it does not exist.
MILVUS_COLLECTION: The name of the Milvus collection. Default is full_text_search. The collection will be created if it does not exist.

Run MCP Server

Use docker compose to start up the server

docker compose up -d

This starts a MCP server with Milvus as the backend. The server will be available at http://localhost:3000.

Interact with MCP server using MCP-Cli

You can also use the mcp-cli command line tool to interact with your MCP server. To use it, run the following command:

npx @wong2/mcp-cli --url http://localhost:3000/mcp

This will connect to the MCP server via streamable HTTP. It will prompt you the list of available tools and allow you to call them with JSON arguments. For example:

$ npx @wong2/mcp-cli --url http://localhost:3000/mcp
✔ Connected, server capabilities: tools
? Pick a primitive ›
❯   tool(insert-multiple) - insert the texts into the collection
    tool(insert) - insert the text into the collection
    tool(search) - search the text in the collection

You can use cursor to navigate the list and select the desired primitive. Then, you will be prompted to enter the arguments for the selected primitive.

For tool(insert-multiple), you can enter a JSON array of strings as the argument. However, MCP-Cli doesn't support it yet. You can use the simpleStreamableHttp example in the @modelcontextprotocol/sdk package to interact with this tool API.
For tool(insert), you can enter a string as the argument to insert a single text string into the collection.
For tool(search), you can enter a string as the argument to search for the text in the collection and return the matching results.

Use simpleStreamableHttp example

You can interact with the MCP server using the simple streamable HTTP client from the MCP typescript-sdk repo: https://github.com/modelcontextprotocol/typescript-sdk.git

Here is the usage:

git clone https://github.com/modelcontextprotocol/typescript-sdk.git
cd typescript-sdk
npm install
npx tsx src/examples/client/simpleStreamableHttp.ts

By default, the client will connect to http://localhost:3000 and you should see the following outputs:

MCP Interactive Client
=====================
Connecting to http://localhost:3000/mcp...
Transport created with session ID: 44dff10f-89bc-4965-b280-e1df3eb56a51
Connected to MCP server

Available commands:
  connect [url]              - Connect to MCP server (default: http://localhost:3000/mcp)
  disconnect                 - Disconnect from server
  terminate-session          - Terminate the current session
  reconnect                  - Reconnect to the server
  list-tools                 - List available tools
  call-tool <name> [args]    - Call a tool with optional JSON arguments
  greet [name]               - Call the greet tool
  multi-greet [name]         - Call the multi-greet tool with notifications
  start-notifications [interval] [count] - Start periodic notifications
  list-prompts               - List available prompts
  get-prompt [name] [args]   - Get a prompt with optional JSON arguments
  list-resources             - List available resources
  help                       - Show this help
  quit                       - Exit the program

>

Then you are in the interactive prompt mode. To connect to different MCP server, just type connect [url] and press enter.

Use the list-tools command to list all available tools. There are 3 available tools:

> list-tools
Available tools:
  - insert-multiple: insert the texts into the collection
  - insert: insert the text into the collection
  - search: search the text in the collection

Use the call-tool <name> command to call a tool with optional JSON arguments. For example, here is how you call the insert-multiple tool to insert multiple texts into the collection:

> call-tool insert-multiple {"texts": ["call-tool <name> [args]: Call a tool with optional JSON arguments", "list-prompts: List available prompts"]}
Calling tool 'insert-multiple' with args: {
  texts: [
    'call-tool <name> [args]: Call a tool with optional JSON arguments',
    'list-prompts: List available prompts'
  ]
}
Tool result:
  2

The number: 2 indicates that the insert operation has inserted two records into the collection.