Updated June 2026 · MyoAmigo
An MCP server for your fitness data
An MCP server for fitness data is a service that exposes a person's real training data — workouts, sets, personal records, volume, body metrics — to any MCP client through a standard set of tools, so an AI assistant can read and act on it without a custom integration. MyoAmigo runs one. It's a hosted, remote MCP server backing a workout tracker for iPhone and Apple Watch, live at api.myoamigo.com/mcp, with 34 account-scoped tools that read history and analysis and write routines, plans, and logged sets.
What it exposes
The Model Context Protocol (MCP) is the open standard Claude, ChatGPT, and Gemini use to reach real tools. MyoAmigo implements the server side of it. The tool surface — the full, current production surface — splits three ways:
- Read: workouts with set-level detail, per-exercise history, personal records, cardio, body metrics, and a dense context snapshot built for agents.
- Analyze: estimated-1RM trends, a PR timeline, stall detection with a concrete prescription, the MyoMap per-muscle weekly heatmap, muscle volume against evidence-based bands, strength standards by bodyweight, honest session verdicts, and a coach-style week-in-review — the same engines the app runs.
- Write: create and revise routines and weekly plans; log sets and cardio. Writes carry idempotency keys, so a retried call never double-logs.
The machine-readable catalog of every tool, with its schema, is at api.myoamigo.com/tools.json.
Transport and authentication
The server is reachable over HTTP at https://api.myoamigo.com/mcp. Clients authenticate one of two ways:
- OAuth 2.1 with Dynamic Client Registration (DCR) and PKCE — clients that support it register and sign in automatically, with no key to paste.
- Personal API key (
myo_sk_…) — for clients or scripts that prefer a static credential.
Every tool call is scoped to the authenticated user's account; a request only ever sees that user's data. Access is revocable at any time. The interactive setup walkthrough is at api.myoamigo.com/connect, and the reference hub is /mcp.
Connecting common clients
| Client | How to connect | |
|---|---|---|
| claude.ai / Desktop | Claude | Add a custom connector, paste the URL; OAuth runs automatically |
| Claude Code | CLI | claude mcp add --transport http myoamigo https://api.myoamigo.com/mcp |
| ChatGPT | Connectors / apps | Add an MCP connector pointed at the server URL, then sign in |
| Other | Any MCP client | OAuth 2.1 (DCR + PKCE) or a myo_sk_… API key |
Step-by-step versions for Claude and ChatGPT live on use Claude as your workout tracker and give ChatGPT your real workout history.
Example calls an agent makes
From a user prompt, the assistant chooses tools and chains them. For instance:
- "Why has my squat stalled?" → read per-exercise history, run stall detection, return the prescription.
- "What should I bench today based on my last 12 sessions?" → read recent bench history and estimated 1RM, propose a working weight.
- "Build a 4-day block and schedule it." → run the plan generator, then write the routines and weekly plan into the app.
- "Log 5 reps at 185 lb on bench." → write a logged set with an idempotency key.
- "How balanced was my week?" → read muscle volume, compare against evidence-based bands, return the heatmap summary.
- "Summarize my training block." → run the week-in-review engine over the range.
Where the data comes from
The underlying log is MyoAmigo, a local-first tracker for iPhone and Apple Watch. It works fully offline with no account required — sets land in an on-device database immediately, and the Apple Watch syncs directly with the iPhone. Optional encrypted cloud sync is what publishes the data to the MCP server so a remote agent can reach it. The catalog has 881 built-in exercises plus custom ones, and Strong CSV import is lossless (kilograms stay kilograms). Get it on the App Store.
Privacy and scope
Data is never used to train AI models and never sold. Access is account-gated on every request, nothing is shared by default, and nothing is shown to another user's agent. You can revoke a client's access at any time.
Related reading
For the concept without the developer framing, see an AI-ready strength training log. Background: what is an MCP server and the AI personal trainer. Full product overview: the AI page.
Frequently asked questions
What is an MCP server for fitness data?
It's a service that exposes a person's real training data — workouts, sets, PRs, volume, body metrics — to any MCP client through standard tools, so an AI assistant can read and act on it without a custom integration. MyoAmigo runs one at api.myoamigo.com/mcp.
How many tools does the MyoAmigo MCP server have?
34, all scoped to the authenticated user's account. They read history, PRs, stats, muscle balance, and week reviews, and write routines, weekly plans, logged sets, and cardio. The catalog is at api.myoamigo.com/tools.json.
How does a client authenticate?
Over OAuth 2.1 with Dynamic Client Registration and PKCE, or with a personal API key (myo_sk_…). Either way, every request is scoped to one account and access is revocable anytime.
Which MCP clients work with it?
Any MCP client — including Claude (custom connector), Claude Code (claude mcp add), ChatGPT (Connectors / apps), and Gemini. The server is a standard remote MCP endpoint, not tied to one vendor.
