apple-health-mcp-server

Install the apple-health-mcp-server, point it at your Apple Health export XML, and register it in your MCP client configuration. The LLM then calls tools like search_records and generate_statistics to answer health questions without SQL.

Set APPLE_HEALTH_EXPORT_PATH to your unzipped export.xml, choose a STORAGE_BACKEND (duckdb, elasticsearch, or clickhouse), then run python -m apple_health_mcp_server.import_data to parse and load the records.

DuckDB is the default local file backend suited for personal use. Elasticsearch fits scalable full-text search workloads, while ClickHouse targets analytics-heavy queries on large datasets.

Yes, it works with any MCP-compatible client including Claude Desktop and Cursor. Add a mcpServers entry with the python command, module arguments, and environment variables to claude_desktop_config.json.

The export ZIP must be unzipped first, and APPLE_HEALTH_EXPORT_PATH must point to the actual export.xml inside the apple_health_export folder. Verify the path with ls and check that the environment variable is set.

DuckDB can slow down on very large exports. Switch STORAGE_BACKEND to ClickHouse for analytics workloads or Elasticsearch for search, and add indexes on record type and start_date columns.