MCP (the Model Context Protocol) is the standard that lets AI assistants use external tools. ChatGPT, Claude and Gemini all speak it now. Chalk is a workout tracker built as an MCP server, which is a fancy way of saying: your AI can operate it.
What the AI can actually do
- Create a full training block: workouts, exercises, sets, reps, target weights, rest times, coaching notes.
- Edit the plan in place when your week changes, without touching the sessions you already logged.
- Log a session for you from a conversation ("log yesterday's squash, 45 minutes, 600 calories").
- Read your recent sessions, including your notes to the coach, and program the next week from real data.
Everything is scoped to your account by a private key baked into your connector link. Revoke the link in Settings and the access dies with it.
The three-minute setup
In Chalk, go to Settings, then Connect your AI coach, and generate your link. In ChatGPT: Settings, Apps and Connectors, enable Developer mode, add a new connector, paste the link, set authentication to none. In Claude: Settings, Connectors, add custom connector, paste the same link. Full click-by-click guides with screenshots of every step live at chalklog.com/guide/chatgpt and chalklog.com/guide/claude.
Design choices that matter
Days are passed by name ("Monday"), not by numeric convention, because language models are much better at names than at remembering whether Sunday is 0. Plan edits happen in place so logged history never detaches from the plan. Deleting a plan never deletes logged sessions. And the tool descriptions teach the model to check what exists before creating anything, which is what stops the classic duplicate-plan mess.
If you are building your own MCP integration, those four decisions will save you real support pain.
Try it
Chalk is free while in early access. Sign up at chalklog.com, connect your assistant, then say: "Build me a 4-week block, 4 days a week, 60 minutes a session." Watch it appear in the app.