VecMed-MCP

VecMed-MCP: Milvus Vector Database for Medical Data

Project Overview

This repository provides tools to establish and manage a Milvus vector database for medical data, specifically designed for rare disease research. It includes scripts for database initialization, data ingestion, search functionality, LLM-based result summarization, and scheduled updates.

File Descriptions

File Name	Purpose
docker-compose.yml	Used to initialize MilvusDB
setup_milvus.py	Formulates the database schema
download_pubmed_tomilvusdb.py	Handles timely updates (currently set to 30 days)
search_milvusdb.py	Tests search functionality in MilvusDB
llm_process_search_result.py	Experiments with summarizing search results using LLM
updating_milvusdb.log	Records database updating operations

ATTU WebUI

The ATTU WebUI provides a visual interface to:

• View all database records
• Manage collections and schemas
• No authentication required

Current Collection: pubmed_rare_disease_db contains over 160,000 PubMed records related to rare diseases.

Environment Configuration

Launch Milvus with Docker Compose

docker compose up -d

Install Required Dependencies

pip install marshmallow==3.20.1 -i https://pypi.tuna.tsinghua.edu.cn/simple
pip install Flask -i https://pypi.tuna.tsinghua.edu.cn/simple
pip install pymilvus -U -i https://pypi.tuna.tsinghua.edu.cn/simple
pip install "mcp[cli]" -i https://pypi.tuna.tsinghua.edu.cn/simple

Set Up ATTU WebUI

docker pull zilliz/attu:v2.4.4
docker run -d --name attu -p 8000:3000 -e MILVUS_URL=192.168.10.199:19530 zilliz/attu:v2.4.4

Workflow

The general workflow can be adapted for various database types beyond medical articles:

1. Build your custom Milvus database with a designed collection schema using the main folder code (steps 1-5)
2. Download required data and store it in your vector database (see download_pubmed_2015-2025 subfolder)
3. Launch the MCP server with HTTP API service or modify to use stdin transport (see pubmed-mcp-server subfolder)
4. Set up timer-based database updates using download_pubmed_to_milvusdb_2.py
5. Integrate the MCP server into your agent/LLM/workflow (example integration with Dify workflow provided)

Scheduled Database Updates

Add a New Cron Job

crontab -e

View Existing Cron Jobs

crontab -l

Author

David Qu
Undergraduate Researcher | AI Algorithm Engineer
University of Toronto Scarborough - Department of Computer Science
📧 davidsz.qu@mail.utoronto.ca