AgentInterOp - Multi-Agent Healthcare Interoperability Platform

A healthcare interoperability platform supporting dual protocols for agent-to-agent communication, real-time FHIR integration, and AI-powered processing.

Features

Core Capabilities

• Dual Protocol Support: A2A (Agent-to-Agent JSON-RPC 2.0) and MCP (Model Context Protocol)
• Real-time FHIR Integration: Live connectivity to FHIR R4 servers for healthcare data
• AI-Powered Processing: Claude integration for narrative-to-JSON conversion and analysis
• Scenario Engine: Pluggable scenarios (BCSE, Clinical Trial, Prior Auth, CQL Measure)
• Agent Card Discovery: Standards-compliant .well-known/agent-card.json implementation

Specialized Agents

• Smart Scheduling Agent: Provider search and appointment booking via SMART Scheduling Links
• Colonoscopy Scheduling Agent: Automates 40+ question intake forms, insurance verification, and appointment booking
• Clinical Informaticist Agent: CQL measure development and quality measure authoring
• Administrator Agent: Benefits eligibility and prior authorization processing
• Applicant Agent: Patient-side conversation handling

Supported Scenarios

Scenario	Endpoint	Description
Smart Scheduling	/api/smart-scheduler/a2a	Provider search and appointment booking
Colonoscopy Scheduling	/api/colonoscopy-scheduler/a2a	Automates complex scheduling workflow
BCSE	/api/bridge/demo/a2a	Breast Cancer Screening Eligibility
CQL Measure	/api/bridge/cql-measure/a2a	Clinical Quality Language measure development
Clinical Trial	/api/bridge/clinical-trial/a2a	Patient enrollment eligibility
Prior Auth	/api/bridge/prior-auth/a2a	Prior authorization processing

Quick Start

1. Install

git clone https://github.com/aks129/AgentInterOp.git
cd AgentInterOp
pip install -e .

2. Configure Environment

Create a .env file:

# Required for AI features
ANTHROPIC_API_KEY=your_anthropic_api_key_here

# Optional
SESSION_SECRET=your_secure_secret_here
PUBLIC_BASE_URL=https://your-domain.com

3. Run

# Development
make run
# or
uvicorn app.main:app --host 0.0.0.0 --port 8000 --reload

# Production
gunicorn --bind 0.0.0.0:8000 app.main:app

4. Access

• Main UI: http://localhost:8000
• Agent 2 Agent Chat: http://localhost:8000/experimental/banterop
• API Docs: http://localhost:8000/docs
• Agent Card: http://localhost:8000/.well-known/agent-card.json

Calling Agents via A2A Protocol

Basic Request

curl -X POST "http://localhost:8000/api/bridge/cql-measure/a2a" \
  -H "Content-Type: application/json" \
  -d '{
    "jsonrpc": "2.0",
    "id": "1",
    "method": "message/send",
    "params": {
      "message": {
        "role": "user",
        "parts": [{"kind": "text", "text": "Build breast cancer screening CQL measure"}]
      }
    }
  }'

Response Format

{
  "jsonrpc": "2.0",
  "id": "1",
  "result": {
    "id": "task-id",
    "status": {"state": "input-required"},
    "history": [
      {"role": "user", "parts": [{"kind": "text", "text": "..."}]},
      {"role": "agent", "parts": [{"kind": "text", "text": "Built CQL measure..."}]}
    ],
    "artifacts": [
      {
        "kind": "file",
        "file": {
          "name": "BreastCancerScreening.cql",
          "mimeType": "text/cql",
          "bytes": "base64-encoded-content"
        }
      }
    ]
  }
}

A2A Methods

Method	Description
message/send	Send message, get response
message/stream	Send message, get SSE stream
tasks/get	Get task status by ID
tasks/resubscribe	Resubscribe to task updates

FHIR Integration

Connect to FHIR R4 servers for patient data:

# Test servers
https://hapi.fhir.org/baseR4
https://r4.smarthealthit.org

Supported operations:

• GET /metadata - Server capabilities
• GET /Patient?name=John - Patient search
• GET /Patient/123/$everything - Complete patient bundle

Development

# Install dev dependencies
make dev

# Code quality
make format     # black, ruff, isort
make lint       # ruff check, flake8
make typecheck  # mypy

# Testing
make test       # pytest
make smoke      # A2A smoke tests

Project Structure

app/
├── main.py              # FastAPI entry point
├── config.py            # Pydantic configuration
├── agents/              # Agent implementations
│   ├── administrator.py
│   ├── applicant.py
│   ├── clinical_informaticist.py
│   └── colonoscopy_scheduler.py
├── protocols/           # A2A, MCP implementations
├── scenarios/           # Healthcare scenarios
├── routers/             # FastAPI route handlers
├── banterop_ui/         # Agent 2 Agent Chat UI backend
├── fhir/                # FHIR server integration
├── llm/                 # Claude AI integration
└── web/                 # Frontend assets
    └── experimental/
        └── banterop/    # Agent 2 Agent Chat UI

api/index.py             # Vercel serverless entry

Deployment

Vercel (Serverless)

Pre-configured for Vercel:

• Routes configured in vercel.json
• Entry point: api/index.py
• Set environment variables in Vercel dashboard

Live Demo: https://agent-inter-op.vercel.app

Docker

docker build -t agentinterop .
docker run -p 8000:8000 -e ANTHROPIC_API_KEY=your_key agentinterop

API Endpoints

Health & Status

Endpoint	Description
GET /healthz	Health check
GET /version	Version info
GET /api/selftest	Self-test suite
GET /.well-known/agent-card.json	Agent discovery

Agent 2 Agent Chat UI API

Endpoint	Description
POST /api/experimental/banterop/scenario/load	Load scenario
POST /api/experimental/banterop/run/start	Start conversation
POST /api/experimental/banterop/a2a/proxy	Proxy A2A requests
GET /api/experimental/banterop/llm/status	Claude status

Environment Variables

Variable	Required	Description
ANTHROPIC_API_KEY	For AI features	Claude API key
SESSION_SECRET	No	Session security (auto-generated)
PUBLIC_BASE_URL	No	Base URL for agent cards
APP_ENV	No	Set to vercel for serverless

License

MIT License - see LICENSE file for details.