Workflow comparison
There is more than one path from Apple Health to an AI workflow.
Manual export is often enough for occasional use. Local or self-hosted apps suit people who want an owned browsing layer. Hosted platforms trade infrastructure control for managed operations. A user-controlled receiver fits recurring, inspectable local agent access.
Comparison at a glance
On a narrow screen, scroll horizontally to compare every column →
| Pattern | Good fit | Main tradeoff | Storage and access |
|---|---|---|---|
| Repeated manual export | Occasional analysis or archival snapshots | A person must initiate and process each new archive | Depends on where the exported files and analysis tools are placed |
| Local app or cache | Owned browsing, organization, or local workflows | Ingestion and agent interfaces vary by project | Can remain local or self-hosted under the operator's controls |
| Hosted platform | Managed infrastructure, APIs, and operational support | The platform's retention, access, export, and account model matter | Data is handled according to the selected service and configuration |
| User-controlled receiver | Recurring delivery to inspectable infrastructure and local agent tools | The user operates a reachable route, receiver, storage, and explicit client setup | Receiver database and bounded read-only access stay on user-controlled infrastructure |
What “Apple Health MCP” means here
Model Context Protocol is a protocol for connecting AI applications to external context and tools. The protocol itself does not fetch Apple Health data. A separate data path must obtain permission, move or store the data, and expose a server interface.
Health Bridge supplies one such path: an iPhone companion reads permitted HealthKit data, sends it to a receiver the user controls, and that receiver provides bounded read-only MCP and CLI interfaces. A compatible local client still needs explicit configuration.
Using Hermes Agent? The Apple Health with Hermes guide documents its same-host stdio registration, tool surface, setup commands, and model-provider privacy boundary.
Using Claude Desktop on macOS? The Apple Health with Claude Desktop guide covers the supported same-Mac setup and manual local server configuration.
Source-linked examples of the patterns
- Manual export: Apple documents a user-initiated XML export from the Health app. The community applehealth project illustrates processing an exported archive.
- Local or self-hosted data app: Fasten Health Onprem illustrates a self-hosted personal health-record application.
- Hosted or developer platform: Medplum illustrates an open-source healthcare developer platform with hosted-service options and APIs.
- User-controlled receiver: the Health Bridge source repository documents its iPhone, receiver, storage, MCP, CLI, and trust boundaries.
Scope of examples: These links illustrate architecture patterns. They are not endorsements and do not assert that the named projects ingest Apple Health, integrate with Health Bridge, or work with a particular AI client.
When Health Bridge is a reasonable fit
- You want recurring delivery rather than repeating an export.
- You are comfortable operating a private route and receiver.
- You want the receiver database on infrastructure you control.
- You want bounded read-only MCP or direct CLI queries on the receiver machine.
- You accept that iOS and network conditions can delay or interrupt delivery.
It is a weaker fit when a one-time export is sufficient, when self-hosting is unwanted, or when a managed platform is a deliberate requirement.
Sources and related pages
- Apple Support: share and export Health data
- Apple Developer: HealthKit authorization
- Model Context Protocol specification
- Category guide, privacy architecture, privacy policy, and support
If the user-controlled receiver pattern matches your needs, use the operational guide for the setup sequence.