Health Auto Export vs MetricBridge: Which Apple Health MCP Server Wins?
Both apps now ship an MCP server, so the question has moved. It is no longer "does this app have one" but "where does it run, and what has to be switched on for your agent to reach it". That single detail picks the winner for your setup.
AI-assisted, human reviewed. How we use AI
The quick answer
Health Auto Export is an export engine that added an MCP server on the phone. Its App Store listing advertises 150+ metrics out to CSV, JSON, Markdown or GPX, and destinations that include iCloud Drive, Google Drive, Dropbox, a REST endpoint, MQTT, Home Assistant and email. The MCP server runs on the iPhone and serves clients on the same local network, and its help pages list Premium access as a prerequisite. MetricBridge is an export engine that ships its MCP server as a file you run on your computer. It exports 194 metrics as JSON or CSV to iCloud Drive, a local folder, your LAN or a webhook, and the open health-export-mcp server reads that export, zero dependencies, no network calls, 16 read-only tools plus two event tools and 22 prompts.
So: if your agent lives on the same Wi-Fi as your phone and you want the shortest possible path from HealthKit to a tool call, Health Auto Export is a reasonable pick and a genuinely well-built exporter. If your agent runs on a laptop, a home server, a container or a hosted runner, MetricBridge is the setup that survives the phone being in another room.
What each one actually is, without the marketing
- Health Auto Export leads with breadth of format and destination. Four formats, seven integration families, plus a Mac and web story. Its MCP server is newer, and its own documentation is honest about the shape: the endpoint is
http://{LAN_IP}:9000/mcpover Streamable HTTP, or TCP on the same port for a Node bridge, and the prerequisites are Premium access, Node 18 or later on the client machine, and both devices on the same network. There are documented configs for Claude Code, Cursor, VS Code and Codex CLI. - MetricBridge leads with how the server is deployed. One Node process, one file, zero dependencies, reading a folder you already have. The iOS app requests read-only HealthKit permission and never writes back, exports 194 metrics as JSON, and keeps a provenance card on every answer it computes in the app.
The three differences that decide it
1. Where the MCP server runs
This is the whole argument. A server on the phone means the phone has to be awake, on the same network, and reachable at a current LAN address. That last part is real friction: iOS hands out new local IP addresses on a schedule, so a client config with a hardcoded address is a thing you get to maintain. Health Auto Export's own help centre has a section on it, which tells you it comes up often.
A server on your computer means the machine your agent already runs on. The phone charges, exports in the background, and your agent reads the last good export whenever it asks. Nothing needs to be reachable except a file.
2. What can reach it
A LAN endpoint is unreachable from a hosted agent, a CI runner, a cloud IDE or a colleague's laptop. If your only client is Claude Desktop on the Mac on your desk, that never bites you. If any part of your setup is not on the same Wi-Fi, it decides the answer on its own.
3. Asking questions in the app, versus only through your agent
Health Auto Export is export-first: the intelligence lives in whatever agent reads the files. MetricBridge also computes answers on the device, with an Ask surface and a weekly brief, and every answer renders as a provenance card: the number, the date range, the sample count, the computation and the source. If you want one place where your agent does all the thinking, export-first is cleaner. If you want to check a number without opening a terminal, the in-app answer is the difference.
Where Health Auto Export is plainly better
No hedging on this: if your downstream needs Markdown or GPX, or an Automations tap into Home Assistant, MQTT, Google Drive or Dropbox, MetricBridge does not currently match it. Our export is JSON or CSV, and our destinations are iCloud Drive, a local folder, your LAN and a webhook. A running coach who wants GPX, or a dashboard that polls MQTT, should stay where the integrations are. The MCP server is one part of the decision, not the whole of it.
Where the difference is measured, not claimed
Metric breadth is the one number both vendors publish themselves: 150+ metrics on the Health Auto Export listing, 194 in the MetricBridge catalogue (the app's own count, and every name is on the metrics page). More is not automatically better, and metrics we do not have are worth less than the three you actually ask about. But it is checkable on both sides, which beats a feature table nobody can audit.
The other checkable claim is the server itself. Our side is one file with zero runtime dependencies, MIT-licensed, and you can watch it read nothing but your export folder:
npx -y health-export-mcp
# env:
# HEALTH_DATA_DIR=/path/to/your/export
# PAIRING_SECRET=the-code-from-the-app
{
"mcpServers": {
"health-export": {
"command": "node",
"args": ["/Users/you/.health-export-mcp/server.mjs"],
"env": {
"HEALTH_DATA_DIR": "/Users/you/Library/Mobile Documents/iCloud~ai~healthexport~app/Documents",
"PAIRING_SECRET": "the-code-from-the-app"
}
}
}
}Without a matching pairing code, every data tool answers Locked rather than an empty result, so a wrong credential never looks like a quiet day of no data. That is the failure mode worth designing out of any health bridge.
That strip is the shape to insist on from either app, or from any agent that repeats a number back to you: the value, the samples, the range, the computation and the source.
Verify it yourself in five minutes
- Export from each app once and count what lands. Formats and metric names are visible immediately, no config needed.
- Point one client at each server. If your agent is not on the phone's network, you already know how the LAN route ends.
- Turn the phone off, or put it on another network for an hour, and ask the same question twice. One of the two keeps answering.
- Check what happens with a wrong credential. A bridge that returns empty data on a bad secret is guessing at you.
The honest summary. Health Auto Export wins on formats and hosted integrations, and its on-device MCP server is a real feature, not a stub. MetricBridge wins on where the server runs, on the number of read-only tools and prompts an agent gets, and on in-app answers with provenance. Pick on transport and destinations first, then on the rest.
An Apple Health MCP server that runs where your agent runs
Export 194 metrics as JSON to a folder you choose, then let any MCP client read it. One file, zero dependencies, no account, and answers computed on the phone.
Where it goes: the app reads Apple Health read-only and writes JSON to a folder you pick. What you get: all 194 metrics, 16 read-only tools and 22 prompts. Price: a 7-day free trial, then $24.99 one time, not a subscription. Also: join the list for the agentic upgrade.
Frequently asked questions
Does Health Auto Export have an MCP server?
Yes. Its help centre documents an MCP server that runs on the iPhone and serves clients on the same local network over Streamable HTTP, at the phone's LAN address on port 9000, with TCP for a Node bridge. Its own help pages list Premium access, a Mac or other device with Node 18 or later, and both devices on the same Wi-Fi as prerequisites, and publish client configs for Claude Code, Cursor, VS Code and Codex CLI.
Which one works with an agent that is not on my home network?
An agent that runs outside your local network cannot reach a server on your iPhone. That is the deciding difference here. The MetricBridge server is a Node process you run wherever your agent runs, and it reads the export folder or iCloud Drive copy, so it works from the last export with the phone anywhere.
Is Health Auto Export free?
Its App Store listing shows a free download with in-app purchases, and its own help pages make Premium access a prerequisite for the integrations and the MCP server. Check the current listing for what each tier includes. This page does not quote its prices because tiers change.
Do either of these send my Apple Health data to a cloud AI?
Neither answers your questions in a vendor cloud on your behalf. Health Auto Export writes files to destinations you configure, and its MCP server serves from the phone on your own network. MetricBridge computes its answers on device and its MCP server makes zero network calls, reading a folder. One honest scope note on our side: the MetricBridge app ships Meta's ad-attribution SDK, so it may send three app events and the advertising identifier after you grant App Tracking Transparency, but no health data ever reaches that SDK.
Can I run both apps at the same time?
Yes. Both read Apple Health through read-only HealthKit permission, so they do not conflict. A common setup is keeping the other app for its Automations into Home Assistant, MQTT or Google Drive while an agent queries MetricBridge's export, because MCP only needs one of them.
MetricBridge is a general-wellness app, not a regulated medical device. It describes and computes from your own data; it does not diagnose, detect, treat or monitor any condition. Wellness information, not medical advice.