Patient Data Access MCP

FastMCP server + Gemini 2.0 Flash LangChain client showcasing RBAC-protected patient data tools backed by SQLite + JWT.

Prerequisites

• Python 3.11+
• GOOGLE_API_KEY for Gemini 2.0 Flash (already referenced inside .env)

Install dependencies:

pip install -r requirements.txt

Environment Variables

.env ships with sample values:

JWT_SECRET – symmetric key used to validate tokens.

DB_URL – path to the SQLite file (sqlite:///app.db by default).

MCP_SERVER_SSE_URL – SSE endpoint exposed by FastMCP (http://127.0.0.1:8000/sse).

GOOGLE_API_KEY – Gemini credential (replace with a valid key).

JWT_* – pre-generated tokens scoped to Doctor, Nurse, LabTechnician, and Admin roles.

Update secrets before using in a real setting.

Running the MCP Server

python server.py --transport sse --host 127.0.0.1 --port 8000

• The first launch creates/ seeds app.db with staff, patients, vitals, labs, etc.
• Other transports are available (stdio, http, streamable-http). Pass --transport stdio to embed inside another process.
• Use python server.py --issue-token Doctor --token-email someone@hospital.test to mint new JWTs.

LangChain + Gemini Client

The client connects to the MCP server through the LangChain MCP Adapter and lets Gemini pick the correct MCP tool per request.

python langchain_client.py --role doctor --query "Order a lipid panel for patient 2"

Flags:

• --role – one of doctor|nurse|labtechnician|adminstaff (selects the JWT token).
• --query – natural-language instruction for Gemini.
• --server-url – override the SSE endpoint if the server runs elsewhere.

RBAC Map ↔ Tools

Every write tool automatically records an audit entry in the audit_log table (who, what, when, payload).