Practical, no-fluff guides for getting your Apple Health data into AI agents, privately.
Apple Health holds years of your most personal data (heart rate variability, sleep stages, workouts, resting heart rate, blood oxygen, and 190 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.
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 →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 →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 →Oura and Whoop data that reaches Apple Health is in your export today, merged by Apple Health. Per-source reconciliation and vendor-score import are in development.
Read the guide →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 →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 →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 →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 →A health server is listed on the Official MCP Registry with three wearable platforms and no Apple Health. What the registry's Apple Health gap tells a buyer.
Read the guide →Row JSON repeats key names on every sample. The worked arithmetic for a 190-metric export, and why the cheapest answer is not loading the file into context at all.
Read the guide →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 →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 →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 →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 →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 →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 →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 16 health tools out of your coding context.
Read the guide →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 →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 →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 →Vaultbeat syncs Apple Health to an encrypted relay and hands your agent 26 tools, 10 of them write. MetricBridge exports a local snapshot to a read-only server that makes zero network calls. Which trust model do you want.
Read the comparison →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 →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 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 →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 →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 →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 →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 →Make export a background job that writes 190 metrics as clean JSON, then feed it live to your agent over a zero-dependency MCP server with 16 read-only tools.
Read the guide →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'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 →Two on-device apps that answer Apple Health questions privately. One is free and stops at chat. The other exports 190 metrics to your own AI agents over MCP.
Read the guide →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 →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 →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 →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 →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 →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 →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 →Four Apple Health MCP servers compared side-by-side: health-export-mcp (zero-dep), the-momentum (DuckDB), Health Auto Export (TCP), and neiltron. Setup times, tools, data models, and privacy.
Read the comparison →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 →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 →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 →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 →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 →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 →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 →Turn HealthKit into clean, AI-ready JSON your agents can actually read, without manual CSV dumps or a cloud account.
Read the guide →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 →Why the giant export.xml and flat CSVs choke LLMs, and how structured JSON with units and metadata wins.
Read the guide →Ask plain-language questions about your sleep, HRV, and steps. Get answers grounded in your real data.
Read the guide →MetricBridge exports 190 Apple Health metrics as clean JSON to the AI agent of your choice: read-only, local-first, no accounts. The MCP server is fully open source.
Download on the App Store View on GitHub