Lingshu FastMCP Medical AI Service

This project implements a FastMCP server for the Lingshu medical AI model and a corresponding client for testing and integration.

Components

1. mcp_server_lingshu.py: FastMCP server wrapping the Lingshu model

2. mcp_client_lingshu.py: Test client demonstrating interaction with the Lingshu FastMCP server

Related MCP server: DICOMweb MCP Server

Server Features

• Medical image analysis

• Structured medical report generation

• Medical Q&A

Prerequisites

• FastMCP framework

• OpenAI API compatible LLM server (e.g., vLLM)

• Required Python packages (install via pip install -r requirements.txt)

Setup

1. Clone the repository

2. Install dependencies: pip install -r requirements.txt

Usage

Use vLLM to serve the Lingshu Model

vllm serve lingshu-medical-mllm/Lingshu-7B  --dtype float16 --api_key api_key --port 8000  --max-model-len 32768

Wrap the server with FastMCP

export LINGSHU_SERVER_URL="http://localhost:8000/v1" 
export LINGSHU_SERVER_API="api_key"
export LINGSHU_MODEL="lingshu-medical-mllm/Lingshu-7B" # the above config depends on your vllm server config
python mcp_server_lingshu.py --host 127.0.0.1 --port 4200 --path /lingshu --log-level info

Try connecting Lingshu with MCP

export LLM_SERVER_URL="xxx"
export LLM_SERVER_API="xxx"
export LLM_MODEL="xxx" ## this is your own model
python mcp_client_lingshu.py  --mcp-url http://127.0.0.1:4200/lingshu # the mcp-url should depend on the mcp server you deployed in the last step