apple-health-mcp: Query Apple Health Data with AI Assistants

Overview

apple-health-mcp connects MCP-compatible AI assistants to Apple Health data exported as CSV. It loads the files locally and provides tools for discovering the available data, running analytical queries, and generating health summaries.

It is intended for people who want to ask questions about their sleep, workouts, activity, or heart data without uploading the source export to a separate service. The server works with a specific CSV export format, not Apple’s native export.xml file.

Key Features

• Exposes schema discovery, analytical queries, and health-summary tools through MCP.
• Uses DuckDB to query the CSV data with SQL.
• Reads export files in place and does not make network requests.
• Supports JSON, CSV, or summary output for analytical queries.
• Can generate weekly, monthly, or custom health summaries.
• Restricts database file access to the configured health data directory and disables network access and temporary disk storage.

Use Cases

• A personal health-data user can ask an MCP-connected assistant questions about trends across an existing export.
• Someone reviewing workout or activity history can query records across the date range available in their export.
• A user exploring sleep or heart measurements can first inspect the schema, then build queries suited to the files they have.
• Developers can connect an MCP-compatible client to a local data source when they want assistant-driven analysis without uploading the original export.

Project Facts

• Language: TypeScript
• License: MIT
• Stars: 570
• Forks: 24
• Topics: apple-health, claude, duckdb, health-data, mcp, mcp-server, model-context-protocol, quantified-self, sql
• Archived: No

Getting Started

Requires Node.js 22 or newer and an export created with Simple Health Export CSV. Configure an MCP client to run:

{
  "command": "npx",
  "args": ["-y", "@neiltron/apple-health-mcp"],
  "env": {
    "HEALTH_DATA_DIR": "/path/to/your/unzipped/health-export"
  }
}

See the README for client configuration and export instructions, and the querying guide for examples.

Considerations

• Only the Simple Health Export CSV layout is supported; native Apple Health export.xml files are not.
• The first request for a table loads its full CSV history. There is no date window, and loaded tables are held in memory.
• The database is rebuilt from CSV files each time the server starts. Persistent incremental importing is not currently available.
• Device overlap can create duplicate-looking measurements, so queries may need to account for sourceName.
• Query safeguards reduce accidental side effects but do not isolate the process. The project advises using a trusted local stdio MCP client, not exposing the server to untrusted network clients.
• Health reports summarize recorded data and are not medical advice. Query results returned to an MCP client may be sent to that client’s model provider.