Healthcare FHIR MCP Server

FHIR R4 (Fast Healthcare Interoperability Resources) MCP server for healthcare AI applications. Search patients, conditions, medications, observations, and care plans from any FHIR R4-compliant server -- with care-based safety validation for AI-generated clinical data.

Built by MEOK AI Labs -- the team behind MEOK AI OS and the Sovereign Temple care membrane framework.

Why this exists

Healthcare AI needs structured access to clinical data, but existing tools lack safety guardrails for AI-generated content. This server validates AI-created observations against physiological safety ranges before they reach the health record. Inspired by MEOK's care membrane framework, it prevents AI from recording impossible vital signs like a heart rate of 500 bpm or a body temperature of 90 degrees Celsius.

Tools

Installation

pip install mcp requests

Configuration

Set the FHIR_SERVER_URL environment variable to your FHIR R4 server endpoint.

Public test servers

For development and testing, you can use these public FHIR R4 servers:

• HAPI FHIR: https://hapi.fhir.org/baseR4

• SMART Health IT: https://r4.smarthealthit.org

Usage

Run the server

FHIR_SERVER_URL=https://hapi.fhir.org/baseR4 python server.py

Claude Desktop config

Add to ~/.claude/claude_desktop_config.json:

{
  "mcpServers": {
    "healthcare-fhir": {
      "command": "python",
      "args": ["/path/to/healthcare-fhir-mcp/server.py"],
      "env": {
        "FHIR_SERVER_URL": "https://hapi.fhir.org/baseR4"
      }
    }
  }
}

Example calls

Search patients:

Tool: search_patients
Input: {"name": "Smith", "count": 5}
Output: {"total": 142, "patients": [{"id": "123", "name": "John Smith", "birthDate": "1980-03-15", ...}]}

Get conditions:

Tool: search_conditions
Input: {"patient_id": "123", "clinical_status": "active"}
Output: {"total": 3, "conditions": [{"display": "Type 2 diabetes", "code": "44054006", ...}]}

Create observation with safety validation:

Tool: create_observation
Input: {"patient_id": "123", "loinc_code": "8867-4", "display_name": "Heart rate", "value": 72, "unit": "bpm"}
Output: {"created": true, "id": "456", "validation": {"valid": true, "safety_check": "passed"}}

Rejected unsafe observation:

Tool: create_observation
Input: {"patient_id": "123", "loinc_code": "8867-4", "display_name": "Heart rate", "value": 500, "unit": "bpm"}
Output: {"created": false, "validation": {"valid": false, "errors": ["SAFETY: Heart rate value 500 is outside safe range (30-220 bpm)"]}}

Care-Based Safety Validation

The care membrane validates AI-generated clinical observations against known physiological ranges:

Values outside these ranges are rejected with a descriptive error. AI-generated observations are also flagged if they use final status (which should be reserved for clinician-verified data).

Security Considerations

• PHI handling: This server connects to FHIR servers that may contain Protected Health Information (PHI). Ensure your deployment complies with HIPAA, GDPR, or applicable regulations.

• Authentication: Use FHIR_AUTH_TOKEN for production FHIR servers. Never use public test servers with real patient data.

• Transport security: Always use HTTPS for FHIR server URLs in production.

• AI-generated data: All observations created through create_observation are validated against safety ranges and default to preliminary status. AI should never set final status on clinical data.

• Audit: Consider enabling your FHIR server's audit logging (AuditEvent resource) for compliance.

• Access control: The MCP server inherits the permissions of the FHIR auth token. Use the principle of least privilege.

• Rate limiting: Built-in rate limiting prevents abuse. Free tier allows 100 calls/day; Pro tier allows 10,000 calls/day.

FHIR R4 Resources Supported

• Patient

• Condition

• MedicationRequest

• Observation

• CarePlan

Additional resource types can be queried through the validate_resource tool for structural validation.

Pricing

Enterprise tier includes a Business Associate Agreement (BAA) for HIPAA-covered entities.

License

MIT