Report Outlines Health Data Paths for AI Agents

The author says the broader goal can be split into three questions: how an AI agent can conveniently obtain data, how it can process the data correctly, and how trustworthy the results are. The September 18 article addresses the first question. The author also acknowledges limited personal devices and energy, having tested only methods for Apple Health and Oura; other vendors are based on public information.

The report divides exercise and health data into three layers. The first includes basic records such as steps, heart rate, sleep periods and workout routes. The second includes algorithm-derived metrics such as readiness, strain and resilience. The third includes diet, training feel and other manual records. These data sit in three types of sources: aggregation platforms such as Apple Health, Health Connect and Google Health; vendor or specialist apps such as Oura, Whoop, Samsung Health, Garmin, Xunji and Hevy; and reports or manual records such as diet, body composition and medical checkups.

The report says vendor apps usually sync the first layer and some second- and third-layer information to platforms, but do not hand over everything. Some fields have no counterpart in Apple Health or Health Connect, and some composite metrics are core selling points of subscription products. Before choosing a method, the report advises users to ask whether the data is already in an aggregation platform and whether the platform's fields are sufficient.

The first path is using data directly inside AI products. ChatGPT and Claude both support Apple Health access in their iOS apps, and eligible users can authorize health data and ask questions. The report says availability depends on region, account and rollout progress, and usually starts in the United States. For users in China without access to a U.S. environment, the Ant Afu app can connect to Apple Health and answer questions, but the author found the features basic. It handles simple sleep and exercise questions but performs poorly on complex analysis, and the product appears more focused on medical report interpretation and AI consultation than on mature health management.

The second path is using existing connectors or relay services. Strava has an official MCP connector for subscribers that lets AI agents get data directly. Xunji offers an API Skill to buyout users under My, Data Export and Import; the API also supports write-back so an agent can write training records into the app. Hevy can generate an API key in Settings, Developer, but requires a Hevy Pro subscription. Freddy is described as a health data relay server that syncs data from multiple platforms to its own server and provides access through MCP or API. The first data source is free, multiple sources require a subscription, and a higher-priced lifetime buyout is also offered. The report warns that convenience comes with data being stored on a third-party server, and users should confirm which fields can be read, where data is stored, and whether a full export is possible after a subscription ends.

The third path is continuous export from aggregation platforms. Apple Health, Health Connect and Google Health all play this role but work differently. For Apple Health, the report describes Shortcuts, third-party apps and self-built tools. Shortcuts can read Apple Health data, filter by date and time, and support incremental export; if the user has a server or database, Shortcuts can send data to the cloud. An existing project, Aimee1129/apple-health-shortcuts-mcp, is cited as a ready-made option for uploading selected data to a cloud database and letting an AI agent read it through MCP. Health Push can export data as JSON. Health Auto Export is more complete: it supports selecting data types, time ranges and export formats, and can send data to a REST API or cloud storage with a premium version. The receiving end can be a computer, NAS, server, Supabase or Cloudflare. The report notes that lifetime buyout is not currently supported in the China region.

For users with a Mac, a developer account and some iOS development knowledge, the report also describes building a personal app to export data or sync it to a server or computer. It cites the-momentum/open_wearables_ios_sdk as an available SDK. The suggested first step is a minimal app that installs on a real device, completes authorization, queries by date and exports JSON. The app should retain time, unit, source and record identifier, then add upload, incremental sync, deduplication and failure retry. The report cautions that development, installation and long-term stable use are different matters, and Apple's free Personal Team signing has limitations.

For Android, the report distinguishes Google Health, Health Connect and Google Fit. Google Health is both an app and a cloud service, renamed from the Fitbit app and completed in May 2026; Fitbit and Pixel Watch data flow directly into it, and it provides the Google Health API v4, successor to the Fit API. Health Connect is an Android system-level data hub that stores data locally and manages authorization so different health apps can read and write the same records; since Android 14 it has been part of the system framework and does not need separate installation. Google Fit is retiring, with official support for the Fit API until the end of 2026 and the app to be merged into Google Health.

The report says Fitbit or Pixel Watch users have data go directly into Google Health, while other brands depend on whether their apps can write the needed data into Health Connect. Public information and online discussion indicate that international versions from OPPO, vivo and Xiaomi, as well as non-China Samsung devices, support writing to Health Connect. For Xiaomi, authorization is available in Mi Fitness under Account, Add Account. The author could not confirm every device without hardware. Some China Samsung devices can enable Health Connect through developer settings and region adjustments, but this depends on the specific system version and is not a universal solution. For Health Connect, Health Sync and the open-source Health Data Export are ready options; Health Sync can export to cloud storage or a server in multiple formats, while Health Data Export supports export to Google Sheets with incremental updates. For custom work, the report points to official Health Connect documentation and the-momentum/open_wearables_android_sdk. Installing a personal Android app is generally more flexible than on iOS, and debugging or installing on one's own phone does not require a store developer account.

The fourth path is calling official APIs or SDKs directly. If a platform cannot provide the needed data, the report says users must return to the source. Vendor APIs, SDKs or official connectors can read data directly from the original service and are usually more suitable for automation.