Claude Now Connects to Apple Health Natively: Cloud Sync vs On-Device MCP
You can ask Claude about your sleep and heart data two ways. Claude's native connector (beta, January 2026) reads your HealthKit categories on the phone and processes them on Anthropic's servers. The on-device path computes the answer on your iPhone and lets any agent you run read the same data from a folder you control. Here is what each one does with your data.
AI-assisted, human reviewed. How we use AI
Disclosure. MetricBridge makes one of the products compared. Third-party facts come from the vendors' own pages, read on 27 October 2026. Updated 27 October 2026: launch date, plan requirement and data-path wording corrected against Anthropic's pages.
Quick answer. Claude's Apple Health connector reads the categories you allow on your iPhone and processes them on Anthropic's servers when you ask a question that needs them. That's a cloud round-trip per question, on a Pro or Max plan, US only, in beta. The alternative is an on-device health AI that answers on your phone, plus a zero-dependency MCP server that lets any agent you run read the same data from a folder you control.
What Claude actually shipped
On 11 January 2026, Anthropic announced Apple Health and Android Health Connect integrations rolling out in beta through the Claude iOS and Android apps, alongside new HealthEx and Function connectors for healthcare (anthropic.com). MacRumors covered the Apple Health rollout on 22 January 2026. OpenAI had announced ChatGPT Health with its own Apple Health connector four days earlier, on 7 January, so Claude was the second big assistant to do this, not the first. The short version: you toggle a connection in the Claude app, pick which categories to share on the standard iOS Health permission screen, and Claude references your HealthKit data in replies. As of 27 October 2026, Anthropic's support page says Health features are in beta, need a Pro or Max plan, and are available in the US only.
That is genuinely useful. A major assistant reads your steps, sleep and heart data without you exporting a file. The trade-off is the part most write-ups skip.
What happens to your data
Claude's connector is read on your iPhone, processed on Anthropic's servers. Anthropic describes it as working through the device's built-in permission system: Claude reads the allowed categories when you ask a question that needs them, and the model does its reasoning in the cloud. Every follow-up question is another round-trip over the network, so a second company, on top of Apple, sees your most personal measurements.
Credit where it is due. Anthropic states that it does not use users' health data to train its models, that Claude can read but cannot write or modify Apple Health entries, and that health information is not saved to Claude's memory by default (support.claude.com, read 27 October 2026). Those are good terms for a cloud product.
For a lot of people that's an acceptable trade for the convenience. The point is that it's a trade, and it's one you should be able to decline without giving up the ability to ask an AI about your health data. That's where the on-device path comes in.
The on-device alternative
On-device health AI runs on your phone. You ask a question in plain English, the answer is computed on the iPhone, and it shows you the numbers it used (a provenance card), so you can check the work against the published methods instead of taking a smooth paragraph on faith. Nothing reaches us or a cloud AI, so there's no account and no sync to a third party. MetricBridge's Ask works this way in four live modes (Guided with no model, Apple's on-device model, a downloaded model of about 400 MB for phones without Apple Intelligence, or the optional Your Mac mode that words one answer through Ollama on your own Mac; see where your data goes when you ask), and the Weekly Brief summarizes your week from the same local data.
Then there's the agent side. Instead of locking your data to one assistant, you export once to a folder you pick (iCloud Drive, a synced folder, your LAN, or a webhook you run) and point a zero-dependency MCP server at it. Any MCP client (Claude, Cursor, opencode, and more) can then query the same 194 metrics through 18 tools (16 read-only). Your data never transits our infrastructure (the export goes only where you point it), and you keep the model you already use.
Cloud sync vs on-device MCP, side by side
| Claude native connector | On-device + MCP | |
|---|---|---|
| Where the model runs | Anthropic's cloud | Your phone (chat), your machine (MCP) |
| Where your data goes | Read on your iPhone, processed on Anthropic's servers | Your phone, and the folder you chose |
| Used for training | No, per Anthropic | No model of ours ever sees it |
| Writes to Apple Health | No (read-only) | No (read-only) |
| Which agents can use it | Claude only | Any MCP client |
| Cost and availability | Pro or Max plan, beta, US only (read 27 October 2026) | $24.99 one time after a 7-day trial; MCP server is open source |
| Agent tools | Built-in connector, no separate tools | 18 (get_trends, compare_periods, correlate_metrics, and more) |
The developer angle is the quiet one. Claude's native connector is a single vendor, single assistant path. If tomorrow you want Cursor to write a script over your sleep data, or opencode to diff two months of HRV, the native connector does nothing for you. An MCP server is the opposite: write the config once, and every MCP-aware client reads the same local file. You're not re-exporting or re-authorizing anything.
Where Claude's connector is stronger
- Zero setup. A toggle in the app you already use. No install, no folder, no config.
- Model size. Claude's largest models reason over your data; an on-device model is small by comparison.
- More sources. HealthEx and Function connectors bring lab results alongside Apple Health.
Which one should you use?
If you want the fastest setup, have a Pro or Max plan in the US, and you're comfortable with your health data being processed on Anthropic's servers, Claude's native connector does exactly what it says. If you'd rather keep your health history on hardware you control, and still want to ask an AI about it, the on-device plus MCP route is the better fit. The two aren't mutually exclusive, but they make opposite bets about where your data belongs.
For most privacy-conscious users, the deciding question is one sentence: do you want a second company processing your heart rate and sleep history, or not. If the answer is not, the on-device route is the one that exists today.
Building the local path yourself? Install the app, turn on export, then paste this into your Claude Desktop config (full client list in the MCP docs; the server is open source on GitHub, npm health-export-mcp 1.8.0):
{
"mcpServers": {
"health-export": {
"command": "npx",
"args": ["-y", "health-export-mcp"],
"env": { "HEALTH_DATA_DIR": "~/Library/Mobile Documents/iCloud~ai~healthexport~app/Documents" }
}
}
}Ask your Apple Health data anything, without the cloud
Answers computed on your iPhone with the numbers shown, plus an open-source MCP server so Claude, Cursor and every other agent you run read the same 194 metrics through 18 tools (16 read-only). No account.
Where it goes: answers are computed on your iPhone; nothing reaches us or a cloud AI. The export goes only where you point it. About two minutes to set up. $24.99 one time after the trial, everything included, never a subscription. Works on any iPhone: a ~400 MB downloaded model covers phones without Apple Intelligence. See the privacy policy for the full data-flow table.
Frequently asked questions
Does Claude's Apple Health connector keep my data on my phone?
No. Claude reads the Apple Health categories you allow through the iOS permission system when you ask a question that needs them, and processes them on Anthropic's servers. Anthropic says it does not use users' health data to train models, Claude cannot write to Apple Health, and health data is not saved to memory by default (support.claude.com, read 27 October 2026). The on-device alternative computes the answer on your phone and nothing reaches a cloud AI.
Can I use Claude's Apple Health connection with Cursor or another agent?
No. Claude's native connector is specific to Claude. An MCP server is the portable path: export once, and Claude, Cursor, opencode, and any other MCP client can query the same local data.
What is the on-device alternative to syncing Apple Health to the cloud?
An on-device health AI answers questions on your phone, and a zero-dependency MCP server lets your own agents read the same data from a file in a folder you chose. Neither sends your health history to a cloud AI; the export goes only where you point it.
When did Claude add Apple Health support?
Anthropic announced Apple Health and Android Health Connect integrations rolling out in beta on 11 January 2026 via the Claude iOS and Android apps, alongside HealthEx and Function connectors. MacRumors covered the Apple Health rollout on 22 January 2026. As of 27 October 2026 the Health features are in beta, need a Pro or Max plan, and are US only.