Let Your AI Agent Read Your Apple Health, Privately

Your AI knows everything about you, except the data that matters most. MetricBridge is the bridge that gives Claude, ChatGPT and Cursor your real Apple Health data from a file on your own machine, with no account and no MetricBridge server in the path.

By MetricBridge · Updated 17 October 2026 · ~8 min read

AI-assisted, human reviewed. How we use AI

Every major AI assistant can now do reasoning, code, and recall in ways that feel close to a colleague. But ask any of them a real question about you, and they draw a blank. "How is my resting heart rate trending?" "Did I sleep better this week than last?" "Which workouts actually move my numbers?" The agent does not have the data, because the data is trapped inside Apple Health.

You have two honest choices for getting it out. The first is the cloud route: upload years of your most personal measurements to a third-party service, create yet another account, and agree to a recurring fee. The second is a local route: the data goes to a file you control and a server on your own computer. This guide is the second one, end to end.

The problem: your AI is blind to you

Apple Health holds 190 metrics, years of history, and the raw material for genuinely useful answers. Your AI agents can read a codebase, a PDF, a spreadsheet. But the one dataset that would let them actually reason about your body is locked away. The health data you already own is invisible to the tools you already use.

That gap is the entire reason a private bridge exists. You should be able to point an agent at your own health data, on hardware you control, without handing that data to anyone else.

How it works, in three steps

1. Export from the app, on device. Install MetricBridge on iOS, grant read-only HealthKit access, and turn on export. It writes a clean JSON file to iCloud Drive or a local sync path. No upload, no account.
2. Add the MCP server, one command. On the computer where your agents run, point a zero-dependency server at that JSON file. It is a single node invocation, no global install, no Docker.
3. Ask your agent anything. The agent calls the server's read-only tools and answers from live data, with the numbers and the provenance shown.

What it looks like in practice

Here is an example exchange, question to answer, with the tool call and the result visible. Your own numbers will differ. This is the moment the agent stops guessing and starts reading you.

You: Call get_mcp_status on my health-export server.
     What data do I have here?

Agent: [calls get_mcp_status]
       MetricBridge bridge active. Data source: iCloud sync.
       190 metrics available. Most recent data: 2026-10-16.
       Total workouts logged: 873.

You: Compare my sleep this week to last week.

Agent: [calls compare_periods → sleep_analysis]
       This week: avg 7.2 h sleep, 2.1 h deep, 4.8 h core, 0.3 h rem.
       Last week: avg 6.8 h sleep, 1.8 h deep, 4.5 h core, 0.3 h rem.
       Deep sleep is up 17%, total sleep up 6%.

No CSV, no copy-paste, no stale screenshot. Each answer is computed from your local export at the moment you ask, and the agent shows the reasoning and the numbers behind it. You can see exactly which metric and date range the answer came from.

Why privacy is the whole point

A health question is among the most personal things you can ask a machine. The bridge is designed so MetricBridge is never in the data path:

• Read-only. HealthKit access is read-only. The app reads, it does not write back to your health record.
• No MetricBridge server. The MCP server makes zero network calls. It reads a local JSON file and talks to your agent over a local pipe.
• Your choice of agent. If you use a cloud AI agent such as Claude or ChatGPT, the numbers it requests go to that provider under its terms. With a local model, they stay on your machine.
• No account. There is nothing to log in to with MetricBridge.
• Provenance on every answer. Every result carries the metric, the date range, and the real coverage, so you can see where an answer came from.

Our privacy commitments are scoped to your health data and to the MCP server. They are stated plainly in the privacy policy, and the server's behavior is open on GitHub.

What is coming next

Apple Health is one source, and a deep one. But most people also wear a ring or a watch that records its own version of the same signals. Today the export holds one value per metric per day, already merged across sources by Apple Health, so you cannot see where devices disagree. We are developing per-source reconciliation for Oura, Whoop and Garmin, plus import of vendor-only scores from their export files. It is not available yet and there is no beta running. The same promise will hold: read-only, no MetricBridge server, no account, and upcoming agent skills included in the existing purchase.

If you want to hear when it is ready, join the early-access list. Joining does not guarantee access.

Related reading

• → Query Apple Health with Any AI Agent (multi-client setup)
• → How to Keep Your Apple Health Data Private When Using AI Agents
• → Connect Apple Health to Claude via MCP
• → Automate Apple Health Export for AI Agents
• → MetricBridge · product overview & setup