Coach Claude
Most AI fitness coaching is generic. It gives you a plan based on your stated goals, then ignores everything that actually happened. Coach Claude is built around the opposite idea: the coaching comes from the data, not from vibes.
Claude reads real workout data directly via MCP — not a summary, not a screenshot. Actual structured data from an open-source Apple Watch training backend: pace, heart rate zones, interval splits, training load over time.
The MCP layer gives Claude direct access to the training database — recent runs, workout splits, heart rate samples, the training queue. It can see what you actually did, not what you said you did. That changes the quality of the coaching significantly.
Per-session notes analyse each run in context: how did today's effort compare to last week? Was that harder-than-expected easy run a sign of fatigue or just a bad day? Heart rate trend analysis flags drift over longer periods, not just individual sessions.
Grounded in real data
Claude sees actual splits, heart rate curves, and training history — not a summary you typed.
The training queue feeds directly to Apple Watch. Claude can suggest workout adjustments, flag when the schedule needs rethinking after a missed session, and push structured workouts to the watch — warm-up, intervals, cool-down — ready to go.
It's not trying to replace a human coach. It's trying to give you the kind of attentive, data-aware feedback a human coach would give if they had access to everything your watch recorded.
Workouts → Watch
Structured sessions queue directly to Apple Watch via the training backend.