Blog

Strava or Apple Health: which one should your AI read?

Strava now has an MCP connector for ChatGPT and Claude, and so does Apple Health, through HealthSync, which I build. Weigh what follows accordingly. To keep it honest, I asked both the same questions about the same workouts, on the same day, with the same account: my first triathlon and an interval session. Here is what each one returned.

Test 1: a triathlon

I raced the Ventouxman sprint triathlon on September 19 with my Apple Watch in multisport mode. Asked for that day, the two connectors returned:

Strava connectorHealthSync
The watch's recordingOne "Workout" of 1:22:44, with 0 m of distance and a speed of zero throughoutOne multisport workout of 1:22:44
DisciplinesNone: no swim, no ride, no runFive legs in order: swim, T1, bike, T2, run
TransitionsAbsentT1 2:47, T2 1:17, each with its heart rate
Per-leg detailHeart rate for the whole race, nothing per disciplinePer leg: distance, heart rate, strokes for the swim, power and running form for the run

That's Strava itself, not only its connector: the Apple Watch's triathlon lands there as a generic workout with no distance and a heart rate curve, nothing else. To get a swim, a ride and a run on my profile, I split the race by hand with an online tool and uploaded three files; the connector then lists those as three activities with no link between them and no transitions. An assistant reading Strava can't tell you that T1 cost a minute and a half more than T2. Here's what it can say when the race is whole, and how to get one GPX per leg without the online tool.

Test 2: an interval session

On October 9 I ran a descending set of 1 km, 800 m, 600 m and 400 m reps. From the Strava connector I requested every stream it offers for that run, cadence included. It returned heart rate, power, speed and distance; no cadence came back. Its list of stream types has no stride length, vertical oscillation or ground contact time at all. HealthSync returned the step count, which gives cadence, and all three of those, for the whole run and for each rep. So I could ask:

Did my running form fall apart on the last reps today?

healthsync · get_workout 19 legs · runningGroundContactTime · runningVerticalOscillation

No, it tightened. Ground contact time went from 221 to 226 ms on the 1 km and 800 m reps to 212 and 217 ms on the 600s, then 207 and 210 ms on the 400s. Vertical oscillation stayed between 9.2 and 9.8 cm on every rep. Shorter contacts at a steady bounce is what faster running looks like when form holds; there's no sign of the late-session shuffle.

It is the question I most want answered at the end of a hard session, and through Strava's connector there's nothing to answer it with.

Test 3: recovery

The Strava connector reads activities, their streams and laps, your profile, zones, gear, clubs and training plan. It has nothing for what happens between workouts: heart rate variability, resting heart rate, sleep, VO2 max. HealthSync syncs those from Apple Health every day, so "am I recovered from Sunday's race?" has an answer.

Where Strava is the better source

Which one should your AI read?

Strava connectorHealthSync
Triathlon as one race with transitionsNoYes
CadenceNot returned in our testYes, from the step count
Stride length, vertical oscillation, ground contact timeNoYes
HRV, resting heart rate, sleep, VO2 maxNoYes
Segments, best efforts, kudosYesNo
Workouts recorded outside Apple HealthYesNo

If your workouts start on an Apple Watch, read them from Apple Health: that's where they're whole. If they start on a Garmin and live on Strava, Strava's connector is the honest choice. And if your plan allows more than one connector, you can add both and ask each for what it does best.

Try it on your own workouts

HealthSync is on TestFlight now and on the App Store soon. Leave an address and we'll send the link.