I Fed My Blood Work to an AI — Here's What It Said

What You'll Learn

• How to connect Garmin, Strava, TrainingPeaks, and gym apps to one AI agent using MCP servers
• A system prompt structure that gives Claude Desktop enough context to coach you
• How to turn a 65-page blood work PDF into personalised nutrition guidance
• Why converting documents to Markdown before adding them to AI context saves tokens and improves output
• The honest limitations of AI coaching — what it can and cannot replace

Timestamps

• 00:00 Cold open — preview clip
• 00:55 Welcome and guest introduction
• 01:27 James previews the episode
• 02:20 "Not a replacement for a human coach"
• 03:00 James's background: 15 years of data tracking
• 05:10 Screen share begins — Claude Desktop setup
• 06:00 System prompt walkthrough
• 07:21 Live demo 1 — multi-data retrospective (sleep, HRV, run critique)
• 09:05 Sleep data pull from Garmin
• 10:05 Run critique from TrainingPeaks
• 12:08 Live demo 2 — session planning from readiness metrics
• 13:46 Live demo 3 — gym programming via Heavy app
• 15:20 AI notices a bicep training gap
• 17:00 Deep research use case
• 18:41 Injury-prevention framing — "You're not ready to run 100K"
• 19:55 Layer 4 — Randox blood work added to context
• 21:18 Privacy note and data trust
• 21:41 Live query: post-workout nutrition from blood markers
• 23:30 Personalised reference ranges discussion
• 25:26 Value summary and key principles
• 26:19 Wrap-up
• 27:21 Post-production bonus: Zwift workout push
• 27:35 Outro

Key Takeaways

• Context is the product — A system prompt with your goals, training history, and blood markers is what separates this from generic ChatGPT advice.
• MCP servers bridge the gap — Claude Desktop connects to Garmin, Strava, TrainingPeaks, Intervals.icu, and Heavy without writing code.
• Convert before you upload — James turned a 65-page blood work PDF into Markdown at 1% of the original size. Less noise, better output.
• AI coaches are honest but not accountable — The agent will critique your training plan but won't chase you when you skip sessions.
• Blood work becomes actionable — Out-of-range markers explained in the context of your lifestyle and training, not just flagged in red.

Tools & Resources Mentioned

• Claude Desktop — AI agent with MCP server support
• Claude Projects — Persistent system prompts and file context
• Garmin — Wearable for sleep, HRV, body battery, resting heart rate
• Strava — Activity tracking
• TrainingPeaks — Structured training plans and analysis
• Intervals.icu — Free training analytics platform
• Heavy — Gym workout logging app with MCP integration
• Randox — Blood testing service
• Zwift — Indoor cycling platform
• MyFitnessPal — Nutrition tracking (mentioned as future integration)

This episode is for informational purposes only and does not constitute medical advice. Always consult a qualified healthcare professional before making changes to your training, nutrition, or supplementation. AI tools are decision-support aids, not replacements for professional medical guidance.