Comorbidity Analysis MCP

Project Description
Comorbidity Analysis MCP is a context-smart medical risk analysis agent that bridges the gap between raw statistical data and clinical reasoning. Our solution features a deep Model Context Protocol (MCP) integration, allowing AI assistants like Claude to better reinforce medical professionals’ assessment and reasoning.

The working prototype, deployed on GCP, executes a sophisticated 4-step reasoning loop: it Extracts ICD-10 codes from natural language, Queries a Neo4j knowledge graph for statistically significant Austrian population comorbidities (Risk Ratios), Verifies findings against live PubMed queries, and Synthesizes an evidence-based medical report. This approach moves beyond simple keyword matching, delivering explainable, high-trust AI that links real-world prevalence data with peer-reviewed literature.

Developer Experience (DX) & Technology Stack
Our DX approach follows a “Protocol-First” philosophy. Instead of building bespoke REST endpoints, we expose our agent logic exclusively through the Model Context Protocol (MCP). This provides a standardized, discoverable interface where tools (like querying the graph or searching PubMed) are instantly available to any MCP-compliant client (e.g., Claude Desktop, IDEs) without additional glue code.

Interface
Model Context Protocol (MCP): Served via Server-Sent Events (SSE) for real-time, streaming interactions.

Specific Technologies & Tools:

LLM Orchestration: langgraph (Stateful Multi-Step Reasoning), langchain
MCP Server: fastmcp (Python-based MCP SDK)
Database (Graph): Neo4j (storing Austrian disease comorbidity networks and Risk Ratios)
External Knowledge: PubMed API (NCBI E-utilities via Bio.Entrez)
LLMs: OpenAI GPT-4o (via langchain_openai)
Hosting & Infrastructure (GCP): GCP: CloudRun + Docker, gcloud tunnel service, CloudBuild, SecretManager

The solution was brainstormed, architected and implemented from ground up during the hackathon.