ChatGPT is good at reasoning over health data and cannot reach Apple Health on its own. It does not speak MCP, so it cannot call a local tool the way Cursor or Claude can. A webhook bridge fixes that, and the key design choice is whose endpoint receives the data.
The flow is one direction: your iPhone POSTs structured JSON to a URL you configure, and you feed that into ChatGPT. The app never sends your data to a server run by the developer. The payload is clean and typed: every metric has a stable lower-case name, an explicit unit, and dates with a UTC offset, so the model never guesses. The same names the MCP server returns.
Three ways to wire it up, from zero code to fully automated: export to a file and drag it into ChatGPT, stand up a fifteen-line local FastAPI receiver on your LAN, or use n8n or Home Assistant if you already run them. Then the privacy distinction. The MCP path keeps the data on your machine with zero network traffic. The webhook path adds a second hop, from your endpoint to OpenAI, and that hop is when the data leaves your control. Use MCP when your agent supports it, the webhook when you specifically want ChatGPT's reasoning.
Full guide, with the real payload, the FastAPI receiver, and a full ChatGPT walkthrough: https://dub.sh/hBVeLDr
Top comments (0)