There's a new technology quietly changing how AI coaching works, and most runners haven't heard of it yet. It's called MCP — Model Context Protocol — and it solves the biggest problem with AI running coaches: they don't know anything about you.
The problem MCP solves
When you ask ChatGPT or Claude for training advice, you get generic answers. "Run your easy runs slowly." "Taper for 2-3 weeks before a race." "Aim for 80% easy, 20% hard."
This advice is technically correct but useless for personalized coaching. The AI doesn't know your current mileage, your recovery status, your race goals, or whether you slept 5 hours or 9 hours last night.
Some runners try to bridge this gap by pasting their Strava data into the chat. It sort of works, but it's tedious, incomplete, and stale the moment you hit enter.
MCP fixes this by letting AI models directly query your fitness data in real time. Instead of you copying data to the AI, the AI reaches out and pulls what it needs.
How MCP works (simply)
Think of MCP like a universal adapter between AI models and data sources. Here's the flow:
- You connect a data source (like Pairform) as an MCP server
- The AI model (ChatGPT, Claude, etc.) can now call tools to query your data
- When you ask a coaching question, the AI automatically fetches relevant data before answering
For example, when you ask "Should I do intervals tomorrow?", the AI might:
- Pull your last 7 days of workouts (mileage, intensity, TSS)
- Check your current training load (CTL, ATL, TSB)
- Look at last night's sleep and recovery score
- Review your upcoming race schedule and goals
- Then answer with all that context
All of this happens automatically — you just ask the question.
What data is available through MCP?
When you connect Pairform as an MCP server, AI coaches can access:
Workouts: Your recent runs with distance, pace, heart rate zones, elevation, TSS, RPE, and notes. Not just the latest — your full training history.
Training load: CTL (fitness), ATL (fatigue), and TSB (form). The AI can see whether you're fresh, building, or overreaching.
Recovery: Sleep duration and stages, HRV, resting heart rate, and recovery scores from WHOOP, Oura, or Garmin.
Body composition: Weight and body fat trends from Withings or manual entry.
Goals: Your active goals (weekly mileage, race targets) and progress toward them.
Race predictions: Estimated race times based on current fitness.
Personal records: Your fastest times at standard distances.
The AI only sees what you've connected and authorized. You control access and can revoke it anytime.
Setting it up
With Claude Desktop
- Create a free Pairform account and connect your devices (Strava, WHOOP, etc.)
- Generate an API key in your Pairform profile
- Add Pairform as an MCP server in Claude Desktop's settings
- Start asking coaching questions — Claude will automatically query your data
With ChatGPT
ChatGPT supports MCP through its GPT actions and MCP Apps. Connect Pairform as a data source and ChatGPT can pull your fitness data during conversations.
With Pairform's built-in coach
If you don't want to configure MCP manually, Pairform has a built-in AI coach powered by Claude that's already connected to your data. Open the chat and start asking questions — no setup required.
Real coaching with MCP vs. without
Here's what the same question looks like with and without data access:
Question: "My hamstring has been tight after runs. Should I take time off?"
Without MCP: "Hamstring tightness can be caused by overtraining, poor warm-up, or muscle imbalances. Rest if it's painful, stretch regularly, and consider seeing a physiotherapist if it persists."
With MCP: "Looking at your data: you've increased weekly mileage from 30 to 42 miles over the past 3 weeks — that's a 40% jump, which is aggressive. Your TSB is -22 and dropping. Your pace on easy runs has been in Zone 3 most of this week, suggesting you're running too hard on recovery days.
The hamstring tightness is likely from the volume spike combined with insufficient easy-day recovery. I'd suggest: drop back to 35 miles this week, ensure all easy runs stay in Zone 2, and add 10 minutes of hip flexor and hamstring mobility after each run. If the tightness resolves within a week, gradually ramp volume back up at no more than 10% per week."
Same question. Dramatically different answer. The second response identifies the probable cause, connects it to specific data points, and provides an actionable, personalized plan.
Why this matters for the future of coaching
MCP is early-stage technology, but the implications for coaching are significant:
Democratized coaching quality: Previously, getting data-informed coaching required a human coach reviewing your training log ($150-400/month). Now any runner with a Strava account and a recovery wearable can get comparable data-driven advice.
Always-on coaching: Your AI coach is available at 5am before your run, at 10pm when you're planning tomorrow, and at mile 18 of your long run if you need a pacing adjustment.
Multi-AI coaching: Since MCP is a standard protocol, you can use the same data with different AI models. Ask Claude for training analysis, ask ChatGPT for a structured plan, use whichever is strongest for each need.
Continuous improvement: As AI models improve, your coaching improves automatically. The data infrastructure (your connected devices and aggregation layer) stays the same — the intelligence on top keeps getting better.
Getting started
The setup takes about 5 minutes:
- Create a free Pairform account — this is your data aggregation layer
- Connect your devices — Strava, WHOOP, Oura, Withings, etc.
- Choose your AI:
- Use Pairform's built-in Claude-powered coach (no extra setup)
- Or connect via MCP to ChatGPT or Claude Desktop for coaching in those platforms
The hardest part isn't the technology — it's trusting that a 5-minute setup can meaningfully improve your training. Try it for a week. Ask the AI a real coaching question. See how the answer compares to generic advice.
Ready to give your AI coach access to your real data? Get started with Pairform — free, 2-minute setup.