K01 Synthetic Health Data

K01 MCP Server

Generate and query clinically realistic synthetic health data directly from your AI workflow.

K01's MCP server gives AI agents and LLM-powered tools access to privacy-preserving synthetic health data generation. Generate FHIR-compliant patient cohorts, query structured EHR data, and compare across FHIR versions — all with built-in differential privacy. The server never handles real patient records: it serves synthetic data from generators that were trained on real records, under differential privacy, inside a separate controlled environment.

This repository contains documentation and configuration for the K01 MCP server. The server implementation is proprietary and runs on K01 infrastructure.

Available Tools

All tools support both FHIR R4 and R5. See docs/tools.md for full parameter documentation.

Quick Start

Claude Desktop

Add to your Claude Desktop configuration (Settings > MCP Servers):

{
  "mcpServers": {
    "k01": {
      "url": "https://mcp.k01.is/mcp",
      "headers": {
        "Authorization": "Bearer YOUR_API_KEY"
      }
    }
  }
}

Cursor

Add to your Cursor MCP configuration (.cursor/mcp.json):

{
  "mcpServers": {
    "k01": {
      "url": "https://mcp.k01.is/mcp",
      "headers": {
        "Authorization": "Bearer YOUR_API_KEY"
      }
    }
  }
}

Restart the application after adding the configuration. The K01 tools will appear in the tool list.

Authentication

The K01 MCP server uses Bearer token authentication. Include your API key in the Authorization header.

To get an API key, contact K01 at k01.is. Self-service key provisioning is coming soon.

See docs/authentication.md for details.

Examples

Generate a diabetic cohort

"Generate 100 patients aged 40-65 with Type 2 diabetes"

The generate_synthetic_cohort tool creates a FHIR Bundle with Patient, Condition, MedicationStatement, Procedure, and Observation resources:

generate_synthetic_cohort(count=100, age_min=40, age_max=65, condition="E11.9", seed=42)

Search for complex patients

"Find patients with 3+ medications and chronic conditions"

search_patients(numberOfMedicinesFrom=3, conditionTypes="chronic", _count=20)

Compare FHIR versions

"How do this patient's medications look in R4 vs R5?"

compare_fhir_versions(patient_id="0101302989", resource_type="Medication")

See docs/examples.md for more workflows.

Privacy and Security

All data this server returns is fully synthetic. No real patient data passes through the server, is stored by it or is referenced by it.

• Differential privacy — noise injection with configurable epsilon values ensures generated distributions cannot be reverse-engineered to identify real individuals
• Real data stays behind the firewall — the generators are trained on real records under differential privacy inside a separate controlled environment; only the trained generators, never the records, are reachable from this server
• FHIR-compliant output — all resources conform to HL7 FHIR R4/R5 specifications, making them safe drop-in replacements for real data in development and testing
• Deterministic seeding — use seeds for reproducible results without persisting any patient data server-side

Links

• K01 Website
• API Documentation
• Tool Reference
• Authentication Guide
• Usage Examples
• FAQ
• MCP Protocol

License

This repository (documentation, configuration, and examples) is licensed under MIT. The K01 server implementation is proprietary.