Curaiva Ai

mcp configured in tools
Agents in promptopinion
curaiva ai mcp in market
CHW Dashboard
Patients Dashboard
Curaiva tools
Doctors Dashboard
markets- Agents(BYO) listed in markets

Inspiration We’ve all seen the flashy demos of AI answering medical questions, but in clinical reality, there is a "Last Mile" gap where raw intelligence fails to become an actionable deliverable. We were inspired by the Agents Assemble vision of the "Endgame"—moving past AI silos and into an era of collaboration. We built Curaiva AI to prove that by using standards like MCP and A2A, we can turn fragmented FHIR data into a "Superpower" for clinicians, helping them conquer the last mile of patient care without the burnout of manual data digging.

What it does:

Curaiva AI is a complete healthcare intelligence ecosystem consisting of three interoperable components:

A "Superpower" (MCP Server): A suite of 9 high-utility tools including triage_patient for crisis detection, generate_chw_priority_queue for community health workers, and create_consultation_brief for physician prep.

view below on marketplace https://app.promptopinion.ai/marketplace/mcp/019dc514-5aaf-779f-9704-e0ec607bbf72

A "Superhero" (Curaiva AI Agent): An intelligent orchestrator that uses the COIN protocol to receive requests from other agents, fetch real-time patient data via FHIR R4, and deliver structured clinical insights. view below on marketplace

https://app.promptopinion.ai/api/workspaces/019dbf4c-12b0-7da9-91db-d43f6ac48809/ai-agents/019dd9d6-d494-7686-8ea6-6830fda42dd4

Specialized Coordinator (HealthConnect Agent): A dedicated agent focused on managing multi-agent handoffs and coordinating Community Health Worker (CHW) workflows. View HealthConnect Coordinator on Marketplace https://app.promptopinion.ai/api/workspaces/019dbf4c-12b0-7da9-91db-d43f6ac48809/ai-agents/019de5d3-9ca0-7d6c-896f-59ecd69bf2fc

SHARP Integration: It natively consumes SHARP Context, meaning it automatically recognizes the current Patient, Practitioner, and Encounter IDs directly from the EHR session without manual input.

How we built it:

The architecture is built for speed and interoperability:

Infrastructure: Node.js & TypeScript for the MCP Server, hosted on Render. Intelligence: A hybrid model approach using Mistral Large for complex clinical reasoning (triage) and Groq (Llama 3.3) for sub-second text generation (briefs). Standards: Fully compliant with the Model Context Protocol (MCP) for tool discovery and Agent-to-Agent (A2A) standards for cross-agent communication. Data: Native FHIR R4 integration, tested against HAPI FHIR servers to ensure real-world data compatibility.

Challenges we ran into: The biggest hurdle was mastering SHARP Context propagation. Mapping session-level EHR tokens to specific MCP tool calls required a deep dive into the Prompt Opinion specs to ensure that when an agent asks "What's the status of this patient?", the MCP server knows which patient and which EHR server to talk to. We also spent significant time fine-tuning our clinical prompts to ensure that the AI-generated triage scores were grounded strictly in the FHIR Observation and Condition data provided.

Accomplishments that we're proud of: Zero-Config Interoperability: Our MCP server is truly "plug-and-play"—as soon as it’s added to the Prompt Opinion Marketplace, all 6 tools are instantly discoverable and usable by any other agent. Multi-Model Orchestration: Successfully balancing the "brain" (Mistral) and the "speed" (Groq) to provide clinical depth without making the user wait. Clinical Utility: Moving beyond "chat" and building tools that generate actual documentation (Consultation Briefs) that doctors can use

What we learned: We learned that the future of healthcare AI isn't about building one giant model that knows everything; it's about composition. By building modular MCP tools, we realized that our triage tool could be used by a "Scheduling Agent," while our medication_adherence tool could be used by a "Pharmacy Agent." The Model Context Protocol is the "glue" that healthcare has been waiting for.

What's next for Curaiva AI: Expanding the Toolbox: Adding tools for ICD-10 coding suggestions and insurance prior-authorization automation. Deep EHR Integration: Moving from public HAPI FHIR servers to private Epic/Cerner sandboxes. A2A Collaboration: Partnering with other hackathon participants to see how their "Superpowers" can work alongside the Curaiva Agent to create a truly unified "Agents Assemble" clinical workspace.

Built With

a2a
express.js
fhir-r4
groq
mcp
mistral-ai
node.js
promptopinion
render
sharp
supabase
typescript
vercel
zod

Created by

I was responsible for designing and building the backend infrastructure for Curaiva AI. My primary goal was to create a clean, reliable, and maintainable foundation that enables the AI agents and MCP tools to function effectively without unnecessary complexity.

We chose Express.js (with TypeScript) as the core framework because it allows us to move fast while keeping the architecture simple and focused on solving real healthcare problems. The backend handles critical responsibilities including:
. Exposing and managing the MCP Server tools
. Secure FHIR R4 data integration and processing
. Agent-to-Agent (A2A) communication support
. Context handling for SHARP integration
. API endpoints that power the Superhero (Curaiva AI Agent) and HealthConnect Agent

By keeping the codebase intentionally simple and well-structured, we were able to rapidly iterate, easily integrate new tools, and ensure the entire system remains focused on clinical utility rather than engineering overhead.

This pragmatic approach helped the team deliver a functional and standards-compliant healthcare intelligence ecosystem within a short timeframe.

Michael Gregory
Lead Architect and Full-Stack Developer.

As the Lead Architect and Developer for Curaiva AI, I was responsible for the end-to-end technical implementation of the project. My primary contributions included:

System Architecture: Designed the interoperable bridge between the MCP Superpower and the A2A Agent orchestrators.
MCP Development: Built the Node.js/TypeScript MCP server from scratch, implementing 6 specialized clinical tools with Zod-validated schemas.
A2A Orchestration: Configured the Agent-to-Agent communication flows using the COIN protocol on Prompt Opinion.
FHIR & SHARP Integration: Engineered the native FHIR R4 data fetching logic and handled SHARP context propagation to ensure seamless EHR session integration.
Intelligence Layer: Conducted the prompt engineering and model selection (Mistral Large & Groq/Llama 3.3) to optimize for both clinical accuracy and sub-second latency.
Full-Stack Execution: Managed the deployment to Render and the integration of the Supabase persistence layer.

James Uchechi Okpara
yusuf habib

Updates

yusuf habib posted an update — May 05, 2026 11:21 AM EDT

Most AI in healthcare feels like a flashy demo, but it often fails when it hits the clinic floor. I’m proud to share my contribution to Curaiva AI, a project built to turn fragmented FHIR data into actionable clinical insights.

I focused primarily on the client-side experience, ensuring that clinicians can interact with our "Superpower" tools seamlessly. I also got my hands dirty on the server side, helping build out the MCP tools that handle everything from crisis detection to physician consultation briefs.

By using the Model Context Protocol (MCP) and SHARP integration, we’ve created a "plug-and-play" ecosystem that reduces clinician burnout and lets doctors focus on what matters: the patient.

Log in or sign up for Devpost to join the conversation.

James Uchechi Okpara started this project — May 05, 2026 10:52 AM EDT

Leave feedback in the comments!

Log in or sign up for Devpost to join the conversation.