EHR-MCP v0.1.0

First public release of the framework-agnostic interoperability protocol
for multi-agent healthcare AI.

⚠️ Pre-release — Epic sandbox validated for read operations.
Not recommended for production clinical environments.

What's Included

MCP Tool Server

• 9 registered MCP tools covering Patient, Condition, MedicationRequest,
  Observation, AllergyIntolerance, Encounter, DiagnosticReport, and raw
  FHIR search
• get_patient_context — single-call full clinical context bundle assembly
• Framework-compatible with LangGraph, CrewAI, LangChain, and AutoGen

FHIR R4 Client

• Async FHIR R4 resource fetching with error normalization
• Vendor detection from FHIR capability statement metadata
• Epic sandbox validated

SMART-on-FHIR Backend Services Auth

• RS384 JWT assertion flow
• System-to-system access — no user session required
• Epic Non-Patient-Facing App compatible

ClinicalContextBundle

• Pydantic v2 typed output schema
• Vendor-normalized — agents receive consistent data regardless of EHR
• Optional plain-language clinical summary for LLM consumption

Known Limitations

• Read-only — write operations (Task, Communication) on roadmap
• Epic sandbox tested; Cerner/Meditech validation pending
• Coverage and Claim tools available via search_fhir only

Inspired By

Masayoshi et al. (2025) — EHR-MCP: Real-world Evaluation of Clinical
Information Retrieval by LLMs via Model Context Protocol
arXiv:2509.15957