Start with the question you came with. Each guide answers it, then points you to the one page that does the job.
Apple Health holds years of your most personal data (heart rate variability, sleep stages, workouts, resting heart rate, blood oxygen, and 194 metrics in all), but it lives locked inside your iPhone, invisible to the AI tools you already use. These guides show you how to bridge that gap: how to export Apple Health data to clean JSON and feed it to large language models like Claude, ChatGPT, and Cursor so an agent can actually reason about your body, spot trends, and answer questions in plain language.
Everything here is built around a local-first, read-only, privacy-first approach. Your HealthKit data never has to pass through a third-party server, and most of these workflows run entirely on hardware you control using the open Model Context Protocol (MCP). Whether you are a quantified-self enthusiast, a biohacker, or just curious what an AI can tell you about your sleep, start with the guide that matches what you want to do.
You already use Claude, Cursor, ChatGPT or a local model. You want it answering from your real Apple Health data, read-only, with nothing to host.
How long does each Apple Health connector take to set up?
The real setup times per route from each project's own docs, what the minutes hide, and a fair way to benchmark the row the compare pages leave off.
Read the guide →What does the self-hosted Apple Health MCP route actually cost me?
The four setup steps in its own docs, where the twenty minutes go, the token that expires in 30 days, and the two-minute route next to it.
Read the guide →Is a hosted health connector actually private?
A remote connector keeps your health data on a machine you do not own; a local MCP server reads a file on yours over stdio. The one config detail that tells you which you have, and how to verify the local claim yourself.
Read the guide →Can my AI agent read my HRV and sleep trends together?
The exact MCP tools, the one-line install, a worked example with the real output, and how to plot week-over-week HRV and sleep from your own local Apple Health export.
Read the guide → Claude · ConnectorsClaude's Apple Health button is missing for me. What now?
Anthropic's built-in Apple Health sharing is beta, US-only and limited to Pro, Max and Ultra. The connector route that works everywhere: export on your iPhone, point a local MCP server at the folder, no country gate and no account in the middle.
Read the guide →What should a read-only Apple Health token actually reach?
Scope a health grant to one destination and read-only tools, keep it out of the write path, GPS routes and raw samples, and revoke it in one step.
Read the guide →How do I connect Apple Health to Claude?
Wire your Apple Health metrics into Claude with a zero-dependency MCP server so it can query your data with live tools.
Read the guide →How do I connect Apple Health to Cursor?
Wire your Apple Health metrics into Cursor with the exact mcp.json config, MCP panel verification, and real queries you can run in Composer.
Read the guide →How do I connect Apple Health to opencode?
The exact opencode.json MCP block for a local Apple Health server, how to verify it in the TUI, and the tool-scoping glob that keeps 22 health tools out of your coding context.
Read the guide →Can ChatGPT read my Apple Health data?
The exact webhook endpoint setup, the JSON payload your iPhone sends, and real ChatGPT prompts that turn your health data into actionable answers. Private: you control the endpoint.
Read the guide →Which agents can read my Apple Health, and how do I set each one up?
One zero-dependency MCP server connects Apple Health to Claude Desktop, Cursor, ChatGPT, Windsurf, Codex CLI, opencode, and more. Real config for every client.
Read the guide →What can I actually ask once my agent has my data?
Ask plain-language questions about your sleep, HRV, and steps. Get answers grounded in your real data.
Read the guide →Can Claude Code analyse my health data without anything leaving my Mac?
Claude Code can write and run real analysis scripts against your Apple Health data (correlations, custom aggregates, charts) with nothing leaving your machine.
Read the guide →Can a local model on Ollama read my Apple Health?
Ollama does not speak MCP, and that is why the tutorials stall. Here is the missing piece: a client that can drive Ollama, with real config for Goose, LM Studio, mcphost and Open WebUI, plus the exact health-export server setup.
Read the walkthrough →What is the shortest private path from Apple Health to my agent?
The local bridge that gives Claude, ChatGPT and Cursor your real Apple Health data with one command, no account, no MetricBridge server. Plus how to join the early-access list for what is coming next.
Read the guide →How do I stop exporting by hand every time?
Make export a background job that writes 194 metrics as clean JSON, then feed it live to your agent over a zero-dependency MCP server with 22 tools, 20 of them read-only.
Read the guide →You want a file of your Apple Health data that a program or a model can read, and you want to know where it goes.
Where should my Apple Health export actually live?
The real headers, Standard Webhooks signature verification, the upsert rules, and the six silent failures that make a working-looking receiver quietly lose weeks of data.
Read the pattern →How do I export Apple Health to JSON for an AI?
Turn HealthKit into clean, AI-ready JSON your agents can actually read, without manual CSV dumps or a cloud account.
Read the guide →Which export route keeps my file on hardware I own?
Three routes to JSON: a browser converter, the free built-in export you convert on your own Mac, or an on-device app. All three give you a file. Only two keep it on hardware you own.
Read the guide →How do I turn export.xml into JSON?
How to convert Apple Health export.xml to JSON. Streaming parsers for 1.8 GB files, Python scripts, browser converters, and the iOS app that skips XML entirely.
Read the guide →My export.xml is gigabytes. What now?
A multi-gigabyte export.xml that no editor or parser will open. Why it grows, the three ways to shrink the problem, and the route that skips the XML entirely.
Read the guide →Why won't my export.xml parse, and how do I fix it?
Apple Health export.xml has been broken since iOS 16: 1.7 GB files crashing Python, R, and Go parsers. Here's every DTD fix and the permanent workaround that skips XML entirely.
Read the guide →JSON, CSV or XML: which should I feed a model?
Why the giant export.xml and flat CSVs choke LLMs, and how structured JSON with units and metadata wins.
Read the guide →JSONL or JSON for an agent on my Mac?
Same bytes, different failure. We cut a 3,200-record snapshot at 100 KB: the JSON array recovered zero records and the JSONL reader kept 1,355. Plus where the HealthMirror app drops its JSONL for a local MCP server.
Read the benchmark →How much of my context window does a full export cost?
Row JSON repeats key names on every sample. The worked arithmetic for a 194-metric export, and why the cheapest answer is not loading the file into context at all.
Read the guide →Why does my export have gaps for days that look fine in Health?
Backfilled samples land behind a date cursor and are never read again. The HealthKit reason, the fix, and the MCP calls that show you the gaps in your own file.
Read the guide →You want answers about your health without handing the data to a cloud AI. What on-device means, what it does not, and where each app's data goes.
How do I keep my health data private while using AI agents?
Does ChatGPT Health send your heart rate to OpenAI? A clear breakdown of what stays local vs what leaves your device when you use AI with your Apple Health data.
Read the guide →Which Apple Health AI apps keep my data on the phone?
Which health AI apps for iPhone keep your Apple Health data private? Six apps compared with data flow analysis, and the one that also bridges to Claude and Cursor via MCP.
Read the guide →Can an on-device AI answer real questions, not just show charts?
A private on-device health AI for iPhone that answers questions about your Apple Health data, without sending your health data to the cloud. Every answer shows its numbers.
Read the guide →ChatGPT Health or on-device AI: what leaves my iPhone?
ChatGPT Health sends your Apple Health data to OpenAI servers. On-device health AI answers the same questions on your phone. The data flow comparison, side by side.
Read the guide →Claude's built-in Apple Health connector or a local MCP server?
Claude's new Apple Health connector syncs your HealthKit data to Anthropic's cloud. What actually leaves your phone, and the on-device MCP alternative that keeps it local.
Read the guide →Why do on-device health apps still need an agent bridge?
On-device health AI keeps your data private. But without an MCP bridge, you can't feed that same data to Claude, Cursor, or any AI agent. So few private apps ship both.
Read the guide →Which on-device health apps let my own agent in?
On-device answers are now table stakes. Agent access is not. The apps we checked, why the second half of the product keeps getting skipped, and the five-minute test for any listing.
Read the guide →Can an app be both on-device and agent-connectable?
On-device and agent-connectable are two separate claims, and most Apple Health AI apps hold only one. The four quadrants, the two tests, and the small overlap that keeps both.
Read the guide →If the model runs on-device, does it still show its working?
A 3B model running on the iPhone is still a generative model. Here is why "on device" is not the same as a number you can trace, with a provenance card you can reproduce.
Read the explainer →What do three AI engines say the best private health app is?
We put the same discovery question to Perplexity, Grok and Exa on 2026-10-04. Two named an app that is both on-device and agent-connectable. Perplexity split the question in half and named nothing that is both.
Read the study →Side by side with named products. MetricBridge makes one of the products compared; competitor facts come from their own pages and carry the date we read them.
Should my agent get raw numbers or a computed answer?
Two 2026 designs side by side. A pass-through tool hands your agent samples and lets the model do the reading. An answer-on-device app computes on the phone and shows a provenance card. When each wins, and why MetricBridge ships both.
Read the comparison →Which Apple Health MCP server should I run?
Four Apple Health MCP servers compared side-by-side: health-export-mcp (zero-dep), the-momentum (DuckDB), Health Auto Export (in-app HTTP server), and neiltron. Setup times, tools, data models, and privacy.
Read the comparison →What are the MCP servers for health data, and who are they for?
Which MCP server should you run to query Apple Health, Oura, Whoop, or Garmin data? Zero-dep vs DuckDB vs cloud-hosted, with the four criteria that matter and the one architecture to avoid.
Read the guide →Metrya or MetricBridge: a recovery score or the numbers behind it?
Metrya scores your recovery and sleep. MetricBridge answers your questions on-device with the numbers behind them and hands your data to your own agents. A dated table, and where each one wins.
Read the comparison →Marrow or MetricBridge: on-phone MCP or a local server?
Marrow serves read-only tools from your iPhone over your LAN, free. MetricBridge writes JSON your agent reads over stdio. Which one your agent can actually reach, and the locked-phone reality neither escapes.
Read the comparison →Solair AI or MetricBridge: on-device Q&A with or without an agent bridge?
Two on-device apps that answer Apple Health questions privately. One is free and stops at chat. The other exports 194 metrics to your own AI agents over MCP.
Read the guide →Vaultbeat or MetricBridge: encrypted cloud sync or a local file?
Vaultbeat syncs Apple Health to an encrypted relay and hands your agent 29 tools, 12 of them write. MetricBridge exports a local snapshot to a server whose tools are read-only apart from a local events file, and which makes no outbound network calls. Which trust model do you want?
Read the comparison →Health Auto Export's built-in MCP server or a local one?
The export incumbent bundled the agent path into the app. What the listing claims, what the help page describes, and how an in-app LAN server differs from a zero-dependency local stdio server.
Read the comparison →the-momentum (DuckDB) or health-export-mcp (zero dependencies)?
Two Apple Health MCP servers, two architectures. One requires Docker + Python + DuckDB + manual XML. The other is zero-dep Node.js with a companion iOS app. We compare setup, tools, data model, and privacy.
Read the comparison →vpetersson's Rust server or MetricBridge for a huge export.xml?
A Rust importer into DuckDB, or a small JSON file the iPhone keeps current. Install, tool surface, freshness, and how each one handles a multi-gigabyte export.
Read the guide →Why does the most-starred server need a second app?
The star-leading Apple Health MCP server cannot read export.xml, so a second app has to produce a CSV first. What the two-app tax costs, and the one-app path.
Read the guide →Which Apple Health MCP server should I install?
One runs its MCP server on the iPhone over the local network, one runs it on your computer against a synced export. Where the server runs decides which fits your agent.
Read the guide →Do I want a vector search over my health data?
A new class of MCP server embeds your Apple Health export into a vector store. What similarity search adds, where it breaks on numbers, and what the second pipeline costs.
Read the guide →iOS 27's new Health app or MetricBridge?
Apple's redesigned Health app adds summaries, a readiness score, and Health Age. MetricBridge shows the numbers behind every answer and adds an MCP bridge to your own agents.
Read the guide →What do Apple's Health Age and readiness score actually compute?
How Apple's readiness score and Health Age work, what inputs they use, what Apple hasn't disclosed, and how MetricBridge shows the numbers behind the answer.
Read the guide →Apple pulled the new Health app from the beta. What can I use today?
The redesigned Apple Intelligence Health app is out of the beta and re-timed to later in 2026. What the removal changes for buyers, and the on-device health AI you can run on your iPhone today.
Read the guide →Gentler Streak has Siri now. Is that an agent bridge?
App Intents wire an app to Siri, Shortcuts and Spotlight. They do not wire it to Claude, ChatGPT or Cursor. The difference between a system bridge and an agent bridge.
Read the guide →Does Bevel have an MCP server?
A published capability list describes what the app does for you, not what your agent can do with your data. The feature surface and the access surface, compared.
Read the guide →What does Bevel's 1.1k-vote MCP request tell me?
The most feature-complete cloud coach has a public, heavily voted request for exactly the bridge a local MCP server gives you today. What the board proves, and what it does not.
Read the guide →Can I see Oura, Whoop and Apple Health in one private picture?
Oura and Whoop data that reaches Apple Health is in your export today, merged by Apple Health. Each reading names the device it came from. Vendor-score import is in development.
Read the guide →What cloud health APIs and hosted connectors charge, and what a one-time unlock changes.
What does an Apple Health API cost in 2026?
The cloud health-data APIs charge $299 to $499 a month plus user caps. Here is every published entry tier annualised, and the version with no processor in the path.
Read the guide →A $4.99-a-month connector or a one-time unlock?
A cloud connector now meters Apple Health reads at 100 requests a day. Here is the arithmetic over a year, against a $24.99 one-time unlock with no server in the path.
Read the guide →Will 100 requests a day be enough?
A hosted health connector meters you by request. A local Apple Health MCP server reads your own file as often as you like. The arithmetic of a week of health questions, and why the cap is the wrong unit.
Read the guide →Does my AI plan let me add an MCP server at all?
Claude Free allows one custom connector. ChatGPT Free and Go allow none. The phone apps allow none. The full plan-eligibility matrix, plus the local path that skips the connector quota entirely.
Read the guide →For people shipping or picking a server: registries, scanners and what agents actually see.
Where do agents actually find MCP servers?
The official MCP Registry is the source of truth, directories are cached copies, and freshness decides which build an agent installs. The registry API you can curl, plus a 60-second audit of your own listing.
Read the guide →Why does a registry health server skip Apple Health?
A health server is listed on the Official MCP Registry with three wearable platforms and no Apple Health. What that gap tells a buyer.
Read the guide →How do trust scanners grade an MCP server?
VerifyMCP and the Ext Ops Panel now score published MCP servers off their install line. One health server scored 0 out of 100. The pre-install checks worth bookmarking.
Read the guide →How can my coaching clients share their own data with me, privately?
For fitness, running and wellness coaches: how a client can let their own AI agent read their Apple Health data today, and the coach workflow in development.
Read the guide →Why do “health export ai” and MetricBridge both show up in search?
How to run two brand names without splitting the SERP: one canonical domain, both names on the key pages, and the SERP evidence behind the approach.
Read the guide →MetricBridge exports all 194 metrics as JSON to a folder you choose, and an open-source MCP server with 22 tools (20 read-only) lets Claude, Cursor or any MCP agent read it. Nothing to host. Ask questions in the app too, answered on the iPhone with the working shown.
Start my 7-day free trial Install the MCP serverThen $24.99 one time, not a subscription. No account. Your health data never reaches us or a cloud AI; the export goes only where you point it. Works on any iPhone; the MCP server is open source. Setup takes about two minutes. Where your data goes · How the numbers are computed