Pairform Running - AI Coachign
I built Pairform because every time I asked ChatGPT or Claude for training advice, I had to go through 20 questions to get to advise that was specific/relavent to me.
"How should I train this week?" gets a pretty generic answer (or loads of follow up questions) when the model can't see that you slept 3 hours last night or haven't run in a month....
Pairform connects your Strava, WHOOP, and smart scale into one dashboard, then exposes everything via REST API and MCP so any AI can query your actual data. Think of it as a personal fitness system of record.
What it does:
- Aggregates data from Strava (workouts), WHOOP (sleep, recovery, HRV), and Withings (weight/body comp), with Garmin, Oura,
Fitbit, and Intervals.icu coming soon - Computes training load (CTL/ATL/TSB), estimates your lactate thresholds, tracks 80/20 intensity distribution, and predicts
race times from your VDOT - Built-in AI coach (Claude) that sees your real numbers — this week's mileage, your current form state, active injuries,
planned workouts, sleep trends - Helps you build comprehensive training plans around race
- REST API with a single /api/snapshot endpoint that returns everything an AI needs in one call
- MCP server so Claude Desktop can tool-call directly into your fitness data
The API angle is what I'm most excited about. You can paste your API key into any AI system and suddenly it knows your TSB is +12 (fresh), you averaged 6.8 hours of sleep this week, and your long run last Sunday was 14 miles at 8:20 pace. Then coaching gets dramatically better.
Stack: Next.js, Supabase, Vercel AI SDK with Claude, deployed on Vercel. The MCP implementation is JSON-RPC over HTTP — one endpoint, no sidecar process.
Pricing: Free for everything — all integrations, full dashboard, API and MCP access, training plans, race strategy. I charge $29/mo only for unlimited AI coaching messages (great if you don't already pay for other frontier LLMs). No data hostage games.
This started as a personal tool because I was pretty tired of the "off the shelf" training plans available out there and figured we've all got access to a pretty expert level coach via the newest AI models. They just didn't have access to our fitness data. I've been using it pretty much daily for my own marathon training and it's become the thing I check before every run!
Happy to answer questions about the MCP implementation, the training load math, or anything else.
