Cura AI - Care and Cure Patient Assistant

A Gemini-powered healthcare agent integrating MongoDB MCP to analyze medical reports, diagnose symptoms, match specialists, and schedule appointments with a live JSON-RPC debugging console.

Comment

Inspiration

Patient portals are traditionally static interfaces where users view past appointments and upload PDF reports without truly understanding them. We wanted to change this by creating a proactive, intelligent agent that bridges the gap between patient data and medical action.

Inspired by the potential of Model Context Protocol (MCP), we built Cura—a Gemini-powered medical agent designed to reason, explain reports, check symptoms, recommend specialists, and directly schedule appointments on behalf of the patient, all while displaying a live, transparent MCP protocol trace.

How We Built It

  • Frontend & Design: We built a high-fidelity patient dashboard using React, TypeScript, and Vite, styled with premium HSL colors, soft glassmorphism, and smooth micro-animations.
  • The Brain (AI): We integrated the official @google/genai SDK to leverage Gemini's advanced reasoning.
  • The Superpowers (MongoDB MCP): We implemented a simulated MongoDB MCP Server interface. When patients converse with Cura, Gemini dynamically plans and invokes MCP tools:
    • mongodb_query_clinical_knowledge: Searches medical guides.
    • mongodb_find_specialist: Locates doctors by symptom/specialty.
    • mongodb_get_medical_history: Accesses patient records.
    • mongodb_book_appointment: Modifies database records to reserve slots.
    • mongodb_log_consultation: Saves chat summaries to the database.
  • Developer Observability: We built an on-screen Live MCP Debugger Console that prints JSON-RPC 2.0 messages (like tools/call and tools/call_result) in real-time.

Challenges We Faced

  1. Asynchronous Function Calling: Coordinating the timing between the user’s chat session, the agent's function calls (such as booking appointments), and updating the dashboard UI in real-time.
  2. Standardizing the Protocol: Designing a frontend console that accurately maps the standard JSON-RPC Model Context Protocol specification for educational and presentation value.

What We Learned

  • How to structure complex multi-step agent plans using Gemini's native tool calling.
  • How MCP standardizes LLM-to-database connections, making it incredibly simple to port database tools between different LLM runtimes.

What's Next for Cura AI

  • Arize Phoenix Integration: Implementing automated evaluation for doctor matching and symptom advice.
  • Fivetran Pipelines: Syncing local clinic files with centralized MongoDB Atlas clusters automatically.

Built With

Share this project:

Updates