freddy
Get started Go to dashboard

The complete guide to connecting wearable data to AI

2026-05-09

You wear a Polar, Oura, WHOOP, or Garmin. You also use Claude, ChatGPT, or Perplexity. You've probably wondered why these two pieces of your life don't talk to each other.

The short answer: they can. The path is just slightly indirect, and once you understand what's happening, the rest is mechanical.

The three pieces

To ask AI a question about your wearable data, three things need to fit together:

1. The wearable. It records the data and exposes it through a developer API. Polar has AccessLink, WHOOP has its developer portal, Oura has its Cloud API, and so on. The data is yours; the wearable just stores and serves it.

2. The AI client. Claude Desktop, ChatGPT, Perplexity, and others all support a protocol called MCP (Model Context Protocol). MCP is the open standard that lets AI assistants pull data from external sources during a conversation.

3. An MCP server in the middle. This is the bridge. It reads from the wearable's API on one side and exposes the data to MCP-speaking AI clients on the other. Without it, the AI client has no way to read your wearable; with it, asking the AI about your sleep is the same as asking it about a document you've shared.

That's all there is. AI on one side, wearable on the other, MCP server in the middle.

What an MCP server actually does

An MCP server is a small program that exposes "tools" to AI clients. Tools are typed function calls the AI can make: list_metrics, query_metrics, get_workout_summary. When you ask the AI a question that needs your data, it calls the appropriate tool and gets back structured information it can reason about.

A health-focused MCP server connects to multiple wearable APIs (because most people own more than one wearable, or move between them over time), normalizes the data into one consistent surface, and serves it through MCP. The AI doesn't need to know which wearable produced which data; it just calls the tools and answers your question.

Why this is suddenly possible

Two things changed in 2024-2025 that made this work:

MCP shipped. Anthropic published the Model Context Protocol in late 2024. By 2025, ChatGPT, Perplexity, and several other AI assistants had implemented it. The protocol is open, so anyone can build a client or a server.

Wearable APIs matured. The major wearable companies (Polar, Oura, WHOOP, Garmin, Withings, Suunto) all have OAuth-based APIs that you can authorize once and pull data from continuously. None of this required hacking; it just required someone to build the bridge.

Three ways to set this up

Option 1: Use a hosted MCP server

The fastest path. A service like Freddy handles all of the OAuth flows, data sync, and MCP serving for you. You sign up, connect your wearables, and paste a personal URL into your AI client. Five minutes, no code.

Cost: typically a small subscription. Freddy is $19/year on the paid tier and free for one wearable.

This is the right choice if you want to start using your data with AI today.

Option 2: Self-host an MCP server

If you'd rather run the server yourself, Open Wearables is a community project that does roughly the same job as a hosted service, on hardware you control. You handle OAuth setup, you store the data, you keep the server running.

Cost: free, plus the time to maintain it.

This is the right choice if you self-host other things in your life and want to add this to the stack.

Option 3: Build it yourself

Every wearable's API is documented and standard OAuth. The MCP spec is also documented. If you're a developer who wants to wire one or two specific data sources into a specific AI assistant, you can write the integration in a weekend.

Cost: developer time. Useful if you're building something specific (an iOS app that shows your wearable data via Apple's on-device AI, say) rather than a general-purpose tool.

What conversations look like once it's set up

Once you've chosen a path and connected your data, asking the AI questions is exactly like asking it about anything else. Some examples that work well:

The AI pulls from the MCP server, reads the relevant metrics, and answers in natural language. You can ask follow-ups. You can ask it to look for correlations. You can paste an article you read about sleep and ask whether your data agrees.

Common questions

Is my data private?

Depends on the path. Hosted services hold a copy of your data on their servers (with whatever security their privacy policy describes). Self-hosted means it stays on your machine. Building it yourself means you decide. None of the wearable APIs share your data with the AI provider directly; the data flows through whatever MCP server you're running.

Can I use this with multiple wearables at once?

Yes. Multi-wearable MCP servers (Freddy is one) read from all your authorized devices and expose the combined surface. The AI can correlate across them.

Does the AI keep a copy of my data?

The AI reads your data during a conversation but doesn't typically retain it after. Each new conversation starts fresh and pulls again. Refer to your specific AI client's privacy documentation.

Does this require a subscription to the AI?

It depends on the client. Claude Desktop's MCP support is available on paid plans. ChatGPT requires a Plus or higher. Perplexity has MCP on Pro. The MCP server itself can be free; the AI client typically has its own pricing.

Where to start

If you have a wearable already, the fastest first step is to sign up for a hosted MCP server, connect your wearable, and paste the resulting URL into Claude or ChatGPT. From there, you can ask any question that involves your data and the AI will pull what it needs to answer.

If you don't have a wearable yet and you're choosing one, the comparison of wearable APIs for AI agents covers which devices have the best developer support. For sleep specifically, the Oura vs WHOOP comparison covers the two most-AI-friendly devices in detail.

Updated 2026-05-09.