Why AI gives generic training advice — and how to fix it
Ask ChatGPT or Claude how to improve your training and you get a reasonable answer — the same reasonable answer anyone would get. 'Progressive overload, compound movements, adequate protein.' It is correct. It is also useless at the level of detail that actually moves the needle.
The problem is not the AI. It is that the AI has no data about you. It does not know that your bench press has stalled at 90 kg for three weeks, that your weekly volume on pulling movements is half what it is on pushing, or that you have run four cardio sessions this month averaging 5:40 per km. Without that context, it cannot do better than generic.
The solution is a direct connection between your workout app and your AI assistant. When the AI can read your actual training log, the conversation changes. 'What should I do next session?' becomes a question the AI can answer with real specificity — because it knows what you did last session.
What SammaGym's AI connector does
SammaGym is a free workout tracking app (web PWA, iOS, and Android) with no subscription and no ads. It tracks strength sessions — sets, reps, weight, rest — cardio runs and rides, timed workouts (AMRAP, EMOM, Tabata, For Time), and strength progress over time.
The MCP connector gives any compatible AI assistant read access to your training data, plus a supervised write path for program proposals. 'Supervised' means the AI can suggest a new training program or modifications to your current one, but nothing changes in your app until you open a 'Proposals' screen and approve each change with a single tap.
The connector is compatible with any MCP-capable AI client. The two most accessible right now are ChatGPT (via developer mode / custom connectors) and Claude (on Pro, Max, or Team plans via Settings → Connectors → add a custom connector). Note that a plan supporting custom connectors is required on the AI side — it is an AI-platform constraint, not a SammaGym one.
What your AI can read: the 6 data tools
When you connect your AI assistant to SammaGym, it gets access to six read tools. Each is a structured data source the AI can call to answer your questions.
The 1RM estimates use the Epley formula (weight × (1 + reps / 30)), one of the most widely cited equations in strength research, originally published in Epley (1985) and referenced in the NSCA Essentials of Strength Training and Conditioning. It gives a useful approximation from any submaximal set — no maximum effort required.
What your AI can propose — under your supervision
Reading your data is the foundation. The connector also gives your AI a write path — but one with a strict supervision model built in.
Your AI can propose a new training program or modifications to your current one. A proposal can include warm-up sets, intensity techniques (rest-pause, cluster sets, drop sets), different rep targets per set (pyramid progressions), timed days (cardio blocks, AMRAP, EMOM, Tabata, For Time), and entirely new custom exercises with the correct muscle group assigned.
None of this applies automatically. Every proposal lands in a 'Proposals' screen inside the SammaGym app. You review it, and approve with one tap. Until you approve, your program is unchanged. If the AI proposes something you disagree with, you reject it — the app does nothing without your explicit confirmation.
- You remain in control. The AI cannot silently overwrite your program.
- You can ask for reasoning. Before approving, ask the AI why it structured the proposal the way it did — it has the data to explain.
- You can iterate. Reject a proposal, ask for adjustments, and approve the revised version.
How to connect: ChatGPT and Claude
Connection on both platforms follows the same MCP pattern. You add the SammaGym connector URL, complete a one-click OAuth login to authorize the AI to access your data, and the tools are live. You can disconnect at any time from within your AI platform's settings.
Connecting via Claude (Pro, Max, or Team plan):
- 1Open Claude and go to Settings → Connectors.
- 2Select Add a custom connector.
- 3Enter the connector URL: https://app.sammagym.com/api/mcp
- 4Complete the OAuth login — sign in to your SammaGym account and authorize Claude to access your training data.
- 5The six SammaGym tools are now available in your Claude conversations. Start a new conversation and ask something like 'What does my training look like this month?'
Connecting via ChatGPT (developer mode / custom connectors):
- 1Open ChatGPT and navigate to the custom connectors or plugins section in your account settings (the exact location may vary as OpenAI updates its interface).
- 2Add a new connector and enter the URL: https://app.sammagym.com/api/mcp
- 3Complete the OAuth login to authorize ChatGPT to read your SammaGym data.
- 4In a conversation, ask ChatGPT to use the SammaGym tools — for example, 'Using my SammaGym data, what muscle groups am I undertraining?'
Privacy: your data, your consent
The connector is built on two privacy foundations: explicit consent and data minimization.
Explicit consent. The OAuth login is not a background process. You actively authorize the AI to access your data, and you can revoke that authorization at any time from within your AI platform's settings. The AI gets access only while you have an active, authorized connection.
Data minimization — GPS coordinates stay on your device. When you log a cardio session, your route coordinates are stored locally and never transmitted to SammaGym's servers or made available through the MCP connector. The AI sees cardio metrics only: distance, pace or speed, and calories. The map of where you run is yours alone. This is a deliberate GDPR-aligned design decision, not a limitation.
- Training history, volume, and program data: readable by the AI with your OAuth consent.
- Strength progress and 1RM estimates: readable by the AI with your OAuth consent.
- Cardio metrics (distance, pace, calories): readable by the AI with your OAuth consent.
- GPS route coordinates: never shared — stays on your device only.
Your training data stays in your SammaGym account. The AI gets a view of it through the connector, scoped to the tools described above, for as long as you keep the connection active.
Sources
- Anthropic — Model Context Protocol Specification, modelcontextprotocol.io, 2024.
- OpenAI — ChatGPT Plugins and Custom Connectors documentation, platform.openai.com, 2025.
- Anthropic — Claude MCP Connectors, support.anthropic.com, 2025.
- Epley B. — Poundage Chart, Boyd Epley Workout, 1985. (Original source for the Epley 1RM formula: 1RM ≈ weight × (1 + reps / 30).)
- National Strength and Conditioning Association — Essentials of Strength Training and Conditioning, 4th edition, Human Kinetics, 2016.