HealthAssistAI

Interoperable, Scalable, and Safe Clinical Decision Support via MCP & A2A

Comment

Inspiration

In both the financial and healthcare sectors, the biggest hurdle to AI adoption is Data Integrity and Security. I wanted to build a solution that proves AI can provide high-value clinical insights (like BMI calculation and vital sign triage) without compromising patient privacy. By utilizing the Model Context Protocol (MCP), HealthAssistAI demonstrates how we can separate "reasoning" from "data," creating a secure bridge between AI agents and structured medical records.

What it does

HealthAssistAI is a specialized healthcare agent built with Google AI Studio and deployed on Google Cloud Run. Clinical Triage: Automatically analyzes synthetic vital signs to flag risks like Hypertension. Health Metrics: Processes synthetic height/weight data to provide instant BMI metrics. Agent-to-Agent (A2A) Ready: It is designed to work in a multi-agent ecosystem, allowing other agents to call its "skills" via standardized handshakes. FHIR-Compliant: Uses the FHIR context extension to ensure it can handle industry-standard healthcare data formats.

How we built it

Backend: Developed using Node.js and Express, implementing the JSON-RPC 2.0 standard for MCP. Cloud Infrastructure: Hosted on Google Cloud Run with a "Scale-to-Zero" architecture, ensuring enterprise-grade scalability with zero idle costs. Integration: Seamlessly integrated into the PromptOpinion.ai platform using custom MCP tool definitions and A2A skill sets. Data Safety: Built and tested exclusively using synthetic patient datasets to ensure 100% compliance with PHI regulations.

Challenges we ran into

The primary challenge was ensuring the server perfectly adhered to the MCP handshake protocols (403 and 404 error handling). Transitioning from a standard web server to a JSON-RPC compliant MCP server required precise handling of headers, CORS, and method discovery.

Accomplishments that we're proud of

I am proud of creating a tool that is Interoperable. By enabling FHIR context and A2A availability, HealthAssistAI isn't just a chatbot—it’s a modular medical "skill" that any other agent in the marketplace can utilize.

What we learned

I gained deep insights into the Model Context Protocol and how it is revolutionizing the way AI agents interact with external data and tools. I also refined my skills in deploying secure, unauthenticated Cloud Run services that remain protected via API-level security.

What's next for HealthAssistAI

The current version of HealthAssistAI is a powerful "Proof of Concept," but the roadmap for an enterprise-grade rollout includes: Expanded Clinical Toolset: Integrating more complex medical calculators, such as the Framingham Risk Score and GFR (Glomerular Filtration Rate) calculators, via additional MCP tools. Multi-Modal Analysis: Utilizing Gemini’s vision capabilities to allow the agent to analyze synthetic medical imaging or handwritten (synthetic) prescriptions. Proactive Health Monitoring: Implementing an orchestrator agent that "pings" HealthAssistAI whenever a patient's vitals are updated in the FHIR context, enabling real-time emergency flagging. Financial-Health Integration: Leveraging my background in the financial sector to build a module that estimates out-of-pocket costs for suggested treatments based on insurance policy data.

Built With

Share this project:

Updates