Bevel Published Its Full “Intelligence” Capability List. An MCP Server Is Not On It.
A published capability list is the most honest thing a health app can put out. You can read exactly what it does. Read Bevel's closely, though, and notice what a thorough list like that still does not include: a way for an AI agent you chose to read your data.
Quick answer
Bevel published its full Intelligence capability list in a help article edited on 30 September 2026, describing the in-app intelligence features across its product (evidence E10 in our weekly market pack, read 2026-10-18). An MCP server is not on that list. That is not a flaw in the list. It is the clearest possible statement of the category split: a capability list describes what the app does for you, and an agent bridge describes what your agent can do with your data. Those are different products, and most health apps ship only the first.
What a capability list proves
A capability list that names every feature is a good sign. It means the vendor knows what it ships, and you can compare it against what you need without reading between lines. Bevel's list covers intelligence features built into the app: the questions it can answer, the patterns it surfaces, and how it presents them.
Everything on it is the app acting as the intelligence layer. The app reads your data, the app reasons about it, the app shows you the result. That is a fine shape for a coaching product.
The question the list cannot answer
It cannot tell you whether your own agent can read your data, because that was never the design. There is no named interface, no tool list, no config block, and no export step in the list, because the list is about features, not access.
So a buyer reading it carefully should come away with two facts: the app is capable, and the app is closed to your agent. Both can be true at once, and for a lot of buyers both are fine.
Feature surface vs access surface
| In-app intelligence (capability list) | Agent bridge (MCP server) | |
|---|---|---|
| What it lists | Features the app performs | Tools an agent can call |
| Who reasons about the data | The app's model | Your model, your prompt |
| Where you see the answer | Inside the app | Inside your agent |
| Raw data access | No | Read-only metric tools |
| Portable to another client | No | Yes |
How we sourced this. The capability list and the absence of an MCP server from it are from the vendor's own help article as read on 2026-10-18 (E10), which the page shows as edited 30 September 2026. Help pages change; check the current article before relying on it.
Why this is the right ask, not a complaint
An in-app intelligence layer and an agent bridge are not competing products, and most apps will only ever ship the first. If you want a coach inside the app, the capability list is exactly what you should read, and a long one is a good sign.
If you want your own agent to answer questions about your data, the capability list will not help you, and that is not a knock. It is a filter. Apply it early and you save yourself the afternoon of installing an app and finding there is no interface to point your agent at.
What to look for instead
Four things, and they show up in a README rather than a feature list:
- A documented interface with a named tool list, such as an MCP server.
- A data path you can inspect, ideally a local file rather than a vendor copy.
- An open-source server, so you can read what it does with your data.
- No account or plan gate between you and your own metrics.
MetricBridge ships all four. An app that exports 190 Apple Health metrics as local JSON, and a zero-dependency stdio MCP server with 16 read-only tools that reads that export.
{
"mcpServers": {
"health-export": {
"command": "node",
"args": ["/Users/you/.health-export-mcp/server.mjs"],
"env": {
"HEALTH_DATA_DIR": "/Users/you/Library/Mobile Documents/iCloud~ai~healthexport~app/Documents"
}
}
}
}Ask the agent to call get_mcp_status, then ask it which nights your sleep stages shifted this month. The intelligence is now your agent's, running on your data, on your machine. Nothing it reports is a diagnosis, just a deviation from your own baseline.
FAQ
Does Bevel have an MCP server?
As of the help article we read on 2026-10-18, Bevel's Intelligence capability list does not include an MCP server or any documented interface for third-party AI agents. The list covers in-app intelligence features only.
Does a long capability list mean better health AI?
It means the app ships more in-app features, which is useful if you want a coach inside the app. It says nothing about whether your own agent can read your data, because that is a different layer with a different interface.
What is the difference between in-app AI and an MCP server?
In-app AI means the app's own model reads your data and shows you an answer inside the app. An MCP server is an interface your own agent calls, so your model and your prompt produce the answer, and the raw metrics are available as read-only tools.
Can I use both an in-app coach and my own agent?
Yes. Nothing stops you using an app with a strong in-app intelligence layer and also exporting your data to a local MCP server for your agent. MetricBridge takes the second path, and it does not require a vendor copy of your data.
Doing this yourself? MetricBridge handles the export step: all 190 metrics as local JSON, with the open-source MCP server (GitHub).
An access surface, not just a feature surface
MetricBridge is a $24.99 one-time unlock: 190 metrics as local JSON, the open-source MCP server with 16 read-only tools, and the on-device Ask and Weekly Brief.