MedMemory

PyPI version Python 3.12+ License: MIT Downloads Glama

Privacy-first personal health record MCP server — store medical documents locally with AES-256 encryption and query with Claude.

MedMemory lets you ingest prescriptions, lab reports, and discharge summaries into an encrypted local database, then ask Claude questions about your health history in natural language. Your data never leaves your machine.

Demo

MedMemory Demo

Features

• 8 MCP tools — medications, lab trends, drug interactions, vaccination status, visit history, allergies, health summary, document ingestion
• Privacy-first — AES-256 encrypted SQLite database, nothing sent to cloud storage
• Gemini Vision — reads handwritten prescriptions and scanned PDFs natively
• Drug interaction checker — real-time OpenFDA API lookup against your current medication list
• WHO vaccination schedule — cross-references your records and flags overdue vaccines
• Companion web UI — 6-page Next.js dashboard at medmemory-ui.vercel.app
• Works with Claude Desktop and Cursor

Quick Start

pip install medmemory-mcp
medmemory-setup

Then add to your Claude Desktop config (~/Library/Application Support/Claude/claude_desktop_config.json):

{
  "mcpServers": {
    "medmemory": {
      "command": "medmemory-server"
    }
  }
}

Restart Claude Desktop. MedMemory is ready.

Try the Hosted Demo

No installation needed — connect to the Railway demo server with synthetic patient data:

{
  "mcpServers": {
    "medmemory-demo": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://web-production-ba446.up.railway.app/sse"]
    }
  }
}

Tools

Tool	Description	Example prompt
ingest_health_document	Parse any medical PDF or image into the DB	"Ingest this prescription: /path/to/rx.pdf"
get_current_medications	Return active medication list	"What medications am I on?"
get_lab_trend	Historical readings for any lab marker	"Show me my HbA1c trend"
get_visit_history	Doctor visits with specialty filter	"What did my cardiologist say?"
get_vaccination_status	Cross-reference WHO schedule, flag gaps	"Am I up to date on vaccines?"
check_drug_interaction	OpenFDA lookup against your med list	"Is Ibuprofen safe with my medications?"
get_allergies	Return recorded allergies	"What allergies do I have on record?"
generate_health_summary	Printable one-pager for new doctors	"Generate my health summary"

Privacy Architecture

What	Where it goes
Your health records	Local encrypted SQLite only
Encryption key	Your machine only, never transmitted
Document text during ingestion	Gemini API (transient, not stored)
Drug name during interaction check	OpenFDA API only
Tool call results	Anthropic (same as any Claude conversation)

Verify the encryption yourself:

xxd ~/medmemory.db | head -3
# Should show random bytes — not "SQLite format 3"

See PRIVACY.md for full details.

Web UI

Live at medmemory-ui.vercel.app

6 pages: Dashboard · Upload · Medications · Lab Results · Timeline · Health Summary

Development

# Clone and install
git clone https://github.com/priyanshugoel24/medmemory-mcp
cd medmemory-mcp
uv sync

# Set up environment
cp .env.example .env
# Add GEMINI_API_KEY, OPENFDA_API_KEY, MEDMEMORY_DB_KEY

# Seed test data
uv run seed_data.py

# Run MCP server in dev mode
uv run mcp dev medmemory/server.py

# Run FastAPI bridge
uv run uvicorn medmemory.api:app --reload --port 8000

# Run Next.js UI
cd ui && npm run dev

Project Structure