Epic Patient API MCP Server

Fake Epic MCP server with file-based patient data and LLM-powered search.

Features

Two-tier data access: List summaries first, then fetch details by ID
15 MCP Tools for querying patient data:
- List tools (return IDs, dates, metadata only): list_patient_allergies, list_patient_medications, list_patient_conditions, list_clinical_notes, list_lab_results, list_vitals, list_procedures
- Detail tools (require IDs, return full data): get_allergy_details, get_medication_details, get_condition_details, get_clinical_note_details, get_lab_result_details, get_vital_details, get_procedure_details
- Attachment tool: get_note_attachment
- Search tool: search_patient_data - Natural language search using LLM
Pluggable LLM: Supports Gemini or OpenRouter for natural language search
File-based storage: Patient data in YAML files under patients/

Install: uv pip install -e .
Configure LLM in .env:
- Gemini (default): GEMINI_API_KEY=... (get from https://aistudio.google.com/app/apikey)
- OpenRouter: LLM_PROVIDER=openrouter, OPENROUTER_API_KEY=..., OPENROUTER_MODEL=anthropic/claude-3.5-sonnet
Run: uv run python -m epic_mcp.server

Server runs on http://localhost:8000/sse

Two-tier access minimizes data transfer:

List to see what's available (IDs + metadata):
```
{"patient_id": "patient_001"}
```
Returns: [{"id": "allergy-1", "allergen": "Penicillin", "severity": "severe", "recorded_date": "2015-03-20"}]
Get details for specific items:
```
{"patient_id": "patient_001", "allergy_id": "allergy-1"}
```
Returns: {"id": "allergy-1", "allergen": "Penicillin", "reaction": "Hives, difficulty breathing", "severity": "severe", "recorded_date": "2015-03-20"}

Edit ~/Library/Application Support/Claude/claude_desktop_config.json:

{"mcpServers": {"epic-patient-api": {"url": "http://localhost:8000/sse"}}}