What is an MCP fitness app?

What MCP is, in plain language

Model Context Protocol (MCP) is an open protocol introduced by Anthropic in 2024. It defines a standard way for an AI assistant to discover, connect to, and use external tools and data sources — things like your calendar, a code editor, or a fitness app. Rather than every tool inventing its own integration format, MCP gives developers one common interface and gives AI clients one common way to use it.

Think of it like a universal adapter. Before MCP, connecting an AI to a new data source meant bespoke code on both sides. With MCP, a compatible server (the "MCP connector") and a compatible client (Claude, ChatGPT, or any AI that supports the protocol) can talk to each other out of the box.

Why MCP exists

AI assistants are most useful when they have accurate, current context about you and your situation. Without a standard way to fetch that context, every integration had to be built from scratch by the app developer and the AI provider, then maintained separately on both sides.

MCP solves this by letting an external application expose a set of tools — read operations, search operations, and optionally write or propose operations — through a standard interface. The AI can then discover those tools automatically and call them on your behalf when you give it a task.

The key design principle is that you stay in control. The AI does not get a master key to your account. It gets exactly the permissions you grant when you authorize the connection, and you can revoke them at any time.

MCP vs a normal API

A regular API is a raw interface built for developers. To use a fitness app's API you would need to read its documentation, write authentication code, craft HTTP requests, parse responses, and handle errors. It is powerful but requires technical knowledge and ongoing maintenance.

MCP is a layer on top of that. It packages an API's capabilities into a discoverable, self-describing format that an AI assistant can understand and use on your behalf — without you writing a single line of code. You simply connect the app in your AI client's settings and start asking questions in natural language.

MCP vs pasting workout data into chat

Many people currently work around the lack of integration by exporting data from their app and pasting it into a chat window. This works, but it has real limitations.

• Manual and repetitive. You have to export and paste every time you want updated analysis.
• Static. The AI sees a snapshot. If you logged a session two hours ago, that data is not there unless you paste again.
• Partial. You can only paste what you remember to include. An MCP connector lets the AI pull exactly the current data it needs, live, when you ask.
• No structure. Free-text exports lose the precise relationships between sets, reps, loads, and dates. An MCP tool returns structured data the AI can reason over accurately.

An MCP connection is not magic — it still depends on the quality of data in the app. But it removes the friction of the manual export loop and gives the AI a consistent, structured view of your training history every single time.

What an MCP fitness app lets you do

Once you connect a fitness app via MCP, you can ask your AI assistant questions and give it tasks that would previously have required either a bespoke integration or a lot of copy-pasting. The AI can pull live data from the app, reason over it, and give you a grounded, specific answer.

With SammaGym's MCP connector, a connected AI can read:

• Your training summary: sessions logged, total volume, volume per muscle group, weekly consistency.
• Your active program: days, exercises, sets, reps, intensity targets.
• Your strength progress: estimated 1RM over time, best sets per exercise, bodyweight trend.
• A specific day's detail: every set logged with load and reps.
• Cardio and run metrics: distance, pace, speed, calories, indoor or outdoor flag.
• Timed workout results: AMRAP, EMOM, Tabata, and For Time rounds and scores.

The AI can also propose a new or modified program — including warm-ups, intensity techniques such as drop sets or pauses, pyramid rep schemes, timed days, and custom exercises. The proposal only takes effect after you review and approve it with a single tap in the app's Proposals screen. Nothing changes in your program without your explicit confirmation.

Is it safe?

Security and privacy are legitimate concerns whenever you connect a third-party service to your personal data. Here is how the MCP model addresses them.

• OAuth consent flow. You authorize access through a standard OAuth flow — the same mechanism used by "Sign in with Google" or "Connect to Strava". The AI client never sees your password.
• Revocable at any time. You can disconnect the connector from your AI client's settings. Access is immediately revoked.
• Data minimization. GPS route coordinates never leave your device. The connector exposes aggregated metrics (distance, pace, calories) but not the raw GPS track, in line with GDPR data minimization principles.
• HTTPS only. All communication between the AI client and the MCP server travels over encrypted HTTPS.
• Supervised writes. The connector does not silently modify your program. Any proposed change surfaces as a Proposal you must approve.

How SammaGym implements MCP

SammaGym exposes a remote MCP connector at https://app.sammagym.com/api/mcp. The server implements the MCP specification and describes its available tools so that any compatible AI client can discover and use them without additional configuration.

To connect, you need an AI plan that supports custom MCP connectors — for example Claude Pro, Claude Max, or Claude Team; or ChatGPT with developer mode enabled. The setup is a one-time, one-click OAuth authorization. After that, you can ask your AI assistant about your training in natural language and it will call the relevant tools automatically.

SammaGym is free — available as a web PWA, on iOS, and on Android, with no subscription required to use the app itself. The MCP connector is part of the same free app.

The connector was built with a clear separation between read operations and write operations. Read tools return current data instantly. The write operations — proposing a new program or a change to your current one — are designed so that the AI drafts the proposal but the app applies it only after your explicit tap. This keeps the human in the loop for any change that affects your actual training.