A comprehensive FHIR (Fast Healthcare Interoperability Resources) server and toolkit designed for healthcare data integration, patient care planning, and clinical decision support. This project provides a universal interface to multiple FHIR servers with advanced AI-powered clinical analysis capabilities.
🏥 Overview
The FHIR Careplan project consists of two main components:
- Universal FHIR MCP Server (
fhir_server.py) - A Model Context Protocol (MCP) server that provides standardized access to multiple FHIR servers - FHIR Tools Library (
fhir_tools.py) - A comprehensive Python library for FHIR data manipulation, analysis, and AI-powered clinical insights
✨ Key Features
🔗 Multi-Server FHIR Integration
- Universal FHIR Interface: Connect to multiple FHIR servers (Epic, Cerner, HAPI, Firely, etc.)
- Vendor-Agnostic: Standardized API regardless of underlying FHIR server implementation
- Real-time Connectivity Testing: Automatic server health checks and diagnostics
- Intelligent Failover: Automatic switching between servers for optimal performance
🤖 AI-Powered Clinical Analysis
- OpenAI Integration: Extract clinical keywords and concepts from free text
- Semantic Mapping: Map clinical terms to standardized FHIR codes
- Similar Patient Matching: Find patients with similar clinical profiles
- Predictive Analytics: Generate care recommendations based on historical data
📊 Comprehensive Patient Data Access
- Complete Patient Records: Demographics, conditions, medications, procedures, encounters
- Vital Signs & Lab Results: Categorized observations with time-series data
- Care Plans & Teams: Treatment plans and healthcare provider information
- Allergies & Procedures: Complete medical history tracking
- Async Operations: Non-blocking I/O for high-performance data access
- Intelligent Caching: Condition codes and frequently accessed data caching
- Batch Processing: Efficient handling of multiple patient records
- Connection Pooling: Optimized HTTP connections for multiple servers
🛠 Installation
Prerequisites
- Python 3.11 or higher
- OpenAI API key (for AI features)
- Access to FHIR servers (local or remote)
Setting Up MCP Server
- Clone the Repository
git clone https://github.com/Kushagra-Dutta/Fhir-MCP.git
cd FHIR-MCP
- Install Dependencies
pip install -r requirements.txt
# or using uv
uv sync
- Configure Environment Variables
# Create .env file
echo "OPENAI_API_KEY=your_openai_api_key" > .env
- Run the FHIR MCP Server
Setting Up Firely Test Database Locally
- Obtain a Firely Server License Key
- Visit Firely Server Trial Page
- Fill out the form to receive a license key via email
- You'll receive the license key and download files (license valid for 7 days)
- Save the license file as
firelyserver-license.json
- Set Up Using Docker
# Pull the Firely server image
docker pull firely/server
# Run the container
# For Windows CMD:
docker run -d -p 9090:4080 --name firely.server -v %CD%/firelyserver-license.json:/app/firelyserver-license.json firely/server
# For PowerShell or macOS/Linux:
docker run -d -p 9090:4080 --name firely.server -v ${PWD}/firelyserver-license.json:/app/firelyserver-license.json firely/server
# Verify container is running
docker ps
- Load Test Data
- Use Postman to load test data bundles
- Create a new PUT request in Postman
- Set request type to raw JSON
- Copy content from test data bundles
- Send requests to base URL (http://localhost:9090)
- Repeat for each data bundle
After completing these steps, you'll have a working test database accessible at http://localhost:9090.
Setting Up MCP Chatbot Client
You can set up the MCP chatbot client either by forking the repository or using Docker. The chatbot will serve as the frontend interface for interacting with the FHIR server.
- Get the Chatbot setup
# Clone the repository
git clone https://github.com/cgoinglove/mcp-client-chatbot.git
cd mcp-client-chatbot
- Install PNPM (if not installed)
- Choose Setup Method:
Docker Compose Setup 🐳
# Install dependencies
pnpm i
# Configure environment variables
# Edit .env file with your API keys (created automatically after pnpm i)
# Start all services including PostgreSQL
pnpm docker-compose:up
Local Setup 🚀
# Install dependencies
pnpm i
# Setup environment variables
pnpm initial:env
# Start PostgreSQL (skip if already running)
pnpm docker:pg
# Run database migrations
pnpm db:migrate
# Start development server
pnpm dev
# Optional: Build & start for production-like testing
pnpm build:local && pnpm start
- Configure Environment Variables
Create/edit
.env file with required API keys:# LLM Provider API Keys (add the ones you plan to use)
OPENAI_API_KEY=****
GOOGLE_GENERATIVE_AI_API_KEY=****
ANTHROPIC_API_KEY=****
XAI_API_KEY=****
OPENROUTER_API_KEY=****
OLLAMA_BASE_URL=http://localhost:11434/api
# Auth Configuration
BETTER_AUTH_SECRET=**** # Generate with: npx @better-auth/cli@latest secret
BETTER_AUTH_URL= # Optional: URL you access the app from
# Database Configuration
POSTGRES_URL=postgres://your_username:your_password@localhost:5432/your_database_name
# MCP Configuration
FILE_BASED_MCP_CONFIG=false
# Optional OAuth Settings (for Google/GitHub login)
GOOGLE_CLIENT_ID=
GOOGLE_CLIENT_SECRET=
GITHUB_CLIENT_ID=
GITHUB_CLIENT_SECRET=
- Connect MCP Server to Chatbot
- Access the chatbot at http://localhost:3000
- Create an account and login
- Go to MCP configuration
- Click "Add Server"
- Copy-paste your
.chatbot-config.json configuration
- Configure System Prompt
- Navigate to
src/app/prompts.ts in the chatbot repository - Replace the default system prompt with your custom prompt from
system_prompt.txt - Example system prompt structure:
export const defaultSystemPrompt = `You are a powerful agentic AI coding assistant.
You have access to a set of tools you can use to answer the user's question.
<communication>
1. Be conversational but professional.
2. Refer to the USER in the second person and yourself in the first person.
3. Format your responses in markdown.
4. NEVER lie or make things up.
</communication>
<tool_calling>
Follow these rules regarding tool calls:
1. ALWAYS follow the tool call schema exactly as specified.
2. NEVER call tools that are not explicitly provided.
3. Only calls tools when they are necessary.
4. Before calling each tool, first explain to the USER why you are calling it.
</tool_calling>
<making_code_changes>
When making code changes:
1. Add all necessary import statements and dependencies
2. NEVER output code to the USER, unless requested
3. Use appropriate code edit tools for implementation
</making_code_changes>
`;
- Customize the prompt sections based on your needs:
<communication>: Define how the AI should interact<tool_calling>: Specify rules for using tools<making_code_changes>: Set guidelines for code modifications
- Save the changes and restart the development server
After completing these steps, your chatbot will be connected to the MCP server and ready to interact with the FHIR database.
🏗 Architecture
Core Components
Server Registry
The system maintains a comprehensive registry of FHIR servers:
- Firely Local (
http://localhost:9090) - Local development server - HAPI R4 (
http://hapi.fhir.org/baseR4) - Public test server - Epic, Cerner, Azure - Enterprise healthcare systems (configurable)
📖 Usage Guide
Basic FHIR Operations
1. Server Management
from fhir_tools import UniversalFhirMcpServer
# Initialize the server
fhir_server = UniversalFhirMcpServer()
await fhir_server.initialize()
# Test server connectivity
result = await fhir_server.test_server_connectivity("firely_local")
print(result)
# Switch to different server
await fhir_server.switch_server("hapi_r4")
2. Patient Search
# Search for patients
search_criteria = {
"family": "Smith",
"given": "John",
"birthdate": "1990-01-01"
}
patients = await fhir_server.find_patient(search_criteria)
3. Patient Data Retrieval
# Get comprehensive patient information
patient_info = await fhir_server.get_comprehensive_patient_info("patient-123")
# Get specific data types
conditions = await fhir_server.get_patient_conditions("patient-123")
medications = await fhir_server.get_patient_medications("patient-123")
vital_signs = await fhir_server.get_vital_signs("patient-123")
lab_results = await fhir_server.get_lab_results("patient-123")
AI-Powered Clinical Analysis
# Extract clinical keywords from free text
clinical_note = "45-year-old female with HER2+ invasive ductal carcinoma, stage IIIA"
keywords = await fhir_server.extract_clinical_keywords(clinical_note)
2. FHIR Code Mapping
# Map clinical terms to FHIR codes
clinical_data = {
"conditions": ["breast cancer", "diabetes"],
"age": 45,
"gender": "female"
}
fhir_codes = await fhir_server.map_to_fhir_codes_fast(clinical_data)
3. Similar Patient Matching
# Find patients with similar clinical profiles
criteria = {
"age": 45,
"gender": "female",
"conditions": ["breast cancer"]
}
similar_patients = await fhir_server.find_similar_patients_simple(criteria)
MCP Server Integration
The FHIR server can be used as an MCP server for integration with AI assistants:
{
"mcpServers": {
"fhir-server": {
"command": "python",
"args": ["fhir_server.py"],
"env": {
"OPENAI_API_KEY": "your-key-here"
}
}
}
}
🔧 Configuration
Server Configuration
The system supports multiple FHIR servers configured in fhir_tools.py:
servers = {
"firely_local": {
"name": "Firely Server Local",
"base_url": "http://localhost:9090",
"version": "R4",
"vendor": "Firely",
"auth_type": "none"
},
"hapi_r4": {
"name": "HAPI FHIR R4 Public",
"base_url": "http://hapi.fhir.org/baseR4",
"version": "R4",
"vendor": "HAPI",
"auth_type": "none"
}
}
Environment Variables
# Required
OPENAI_API_KEY=your_openai_api_key
# Optional
FHIR_SERVER_URL=http://localhost:9090
FHIR_SERVER_AUTH_TOKEN=your_auth_token
Core FHIR Operations
switch_server(server_name) - Switch between FHIR serverstest_server_connectivity(server_name) - Test server connectivityfind_patient(search_criteria) - Search for patientsget_comprehensive_patient_info(patient_id) - Get complete patient data
Clinical Data Access
get_patient_observations(patient_id) - Get patient observationsget_patient_conditions(patient_id) - Get patient conditionsget_patient_medications(patient_id) - Get patient medicationsget_vital_signs(patient_id) - Get vital signsget_lab_results(patient_id) - Get laboratory resultsget_patient_encounters(patient_id) - Get patient encountersget_patient_allergies(patient_id) - Get patient allergiesget_patient_procedures(patient_id) - Get patient procedures
AI-Powered Analysis
extract_clinical_keywords(text) - Extract clinical information from textmap_to_fhir_codes_fast(clinical_data) - Map terms to FHIR codesfind_similar_patients_simple(criteria) - Find similar patientsextract_condition_codes_from_fhir() - Extract all condition codes
System Management
list_available_servers() - List all configured serversget_server_registry() - Get complete server registrydiagnose_fhir_server(server_name) - Diagnose server capabilitiesclear_condition_cache() - Clear condition codes cacheget_condition_cache_stats() - Get cache performance statistics
MCP Inspector
The MCP (Model Context Protocol) Inspector is a powerful development and testing tool that helps debug and validate FHIR server interactions. It provides real-time inspection of server behavior and API responses.
Using MCP Inspector
# Run MCP Inspector on a specific file
mcp dev fhir_server.py
# This will:
# 1. Start the inspector in development mode
# 2. Monitor all FHIR server interactions
# 3. Log detailed information to logs/mcp_dev_inspector.log
Features
- Real-time Monitoring: Watch FHIR server interactions as they happen
- Request/Response Logging: Detailed logs of all API calls and responses
- Error Detection: Immediate feedback on API errors or misconfigurations
- Performance Metrics: Track response times and server performance
- Debug Mode: Enhanced logging for development troubleshooting
Log File Location
The inspector writes detailed logs to:
logs/mcp_dev_inspector.log
Best Practices
- Use MCP Inspector during development to validate server behavior
- Monitor the log file for unexpected errors or performance issues
- Run the inspector when implementing new FHIR endpoints
- Use it to debug connection issues with external FHIR servers
🏥 Business Applications
This toolkit can power automated hospital engagement platforms:
- Patient Care Coordination
- Automated care plan generation
- Treatment timeline management
- Multi-disciplinary team coordination
- Clinical Decision Support
- Evidence-based treatment recommendations
- Risk assessment and early warnings
- Outcome prediction based on similar cases
- Resource Optimization
- Predictive analytics for resource allocation
- Automated scheduling and capacity management
- Cost optimization through data-driven insights
- Quality Improvement
- Automated quality metrics tracking
- Compliance monitoring and reporting
- Performance analytics and benchmarking
🔐 Security & Compliance
- HIPAA Compliance: Designed with healthcare data privacy in mind
- Secure Communication: HTTPS/TLS encryption for all server communications
- Authentication Support: Multiple authentication methods (OAuth, API keys, etc.)
- Audit Logging: Comprehensive logging for compliance and debugging
- Async Operations: Non-blocking I/O for high throughput
- Connection Pooling: Efficient HTTP connection management
- Intelligent Caching: Condition codes and metadata caching
- Batch Processing: Efficient handling of multiple records
- Error Handling: Robust error handling and retry mechanisms
🧪 Testing
Running Tests
# Run basic connectivity tests
python -c "
import asyncio
from fhir_tools import UniversalFhirMcpServer
async def test():
server = UniversalFhirMcpServer()
await server.initialize()
result = await server.test_server_connectivity('firely_local')
print(result)
asyncio.run(test())
"
Server Diagnostics
# Comprehensive server diagnostics
diagnostics = await fhir_server.diagnose_fhir_server("firely_local")
print(diagnostics)
📝 Logging
The system provides comprehensive logging:
import logging
# Configure logging
logging.basicConfig(level=logging.INFO)
logger = logging.getLogger(__name__)
# Logs are automatically written to logs/mcp_dev_inspector.log
🤝 Contributing
- Fork the repository
- Create a feature branch
- Make your changes
- Add tests for new functionality
- Submit a pull request
📄 License
This project is licensed under the MIT License - see the LICENSE file for details.
🆘 Support
For support and questions:
- Create an issue in the GitHub repository
- Check the logs directory for detailed error information
- Use the diagnostic tools for server troubleshooting
🔄 Version History
- v0.1.0 - Initial release with core FHIR functionality
- v0.2.0 - Added AI-powered clinical analysis
- v0.3.0 - Enhanced multi-server support and caching
🎯 Roadmap
🕒 FHIR Timeline Agent
The FHIR Timeline Agent (fhir_timeline_agent.py) is a specialized agent designed to generate detailed clinical treatment timelines from patient queries using real FHIR data. It's specifically configured to work with the Firely Local FHIR server.
Features
- Natural Language Processing: Converts free-text patient queries into structured clinical data
- Real Patient Data Analysis: Uses actual FHIR patient records for timeline generation
- AI-Powered Timeline Generation: Leverages OpenAI GPT-4 for accurate clinical timelines
- Interactive CLI Interface: User-friendly command-line interface with rich formatting
- Comprehensive Patient Matching: Finds similar patients based on age, gender, and conditions
Usage
# Run in interactive mode with chat interface
python fhir_timeline_agent.py
# Run in demo mode with example queries
python fhir_timeline_agent.py --demo
Example Queries
• "45-year-old male with pancreatic adenocarcinoma"
• "62-year-old female with HER2+ breast cancer"
• "58-year-old male with stage IIIA lung adenocarcinoma"
Timeline Generation Process
- Query Processing
- Extracts clinical keywords from natural language
- Identifies age, gender, conditions, stage, and biomarkers
- FHIR Code Mapping
- Maps clinical terms to standardized FHIR codes
- Uses Firely Local server's code systems
- Similar Patient Search
- Finds matching patients in the FHIR database
- Scores matches based on age, gender, and conditions
- Data Aggregation
- Collects comprehensive medical history
- Includes procedures, medications, encounters
- Timeline Generation
- Creates detailed treatment timeline using AI
- Organizes events chronologically with clinical context
The agent generates rich, formatted output including:
- Patient Profile: Demographics, diagnosis, stage, biomarkers
- Treatment Timeline: Step-by-step clinical events with dates
- Clinical Outcomes: Treatment response, survival status, toxicity
- Data Sources: Server information and analysis metrics
Configuration
The agent is hardcoded to use:
- Firely Local FHIR server (
http://localhost:9090) - OpenAI GPT-4 for timeline generation
- Rich console output for formatted display
Requirements
- OpenAI API key (set in
.env file) - Running Firely Local FHIR server
- Python packages:
openai>=1.86.0
rich>=13.7.0
python-dotenv>=1.0.0
aiohttp>=3.8.0
Best Practices
- Provide complete patient information in queries
- Include age, gender, and primary diagnosis
- Add stage and biomarker information when available
- Use specific clinical terms for better matching
Error Handling
The agent includes robust error handling for:
- Missing patient information
- FHIR server connectivity issues
- AI generation failures
- Data parsing errors
Each error is displayed with helpful suggestions for resolution.
Built with ❤️ for healthcare interoperability and patient care improvement