Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type @ followed by the MCP server name and your instructions, e.g., "@TrackLab MCP Ultimate Edition What's my training load and recovery readiness today?"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
Enterprise-Grade Model Context Protocol (MCP) Server & AI Running Engine for Claude, Cursor, Windsurf, VS Code, and Gemini CLI.
TrackLab MCP turns any AI Assistant into a physiological AI Running Coach. Built with a plugin architecture, it abstracts raw fitness APIs (Intervals.icu, Garmin, Strava, COROS, TrainingPeaks) and exposes 70+ structured analytical MCP tools powered by literature-backed endurance models (Daniels VDOT, Norwegian Double Threshold, Casado, Seiler TID, Riegel Predictions, ACWR Injury Risk).
🔌 Plugin Architecture Layer: Built with provider abstraction (ProviderInterface). Defaults to Intervals.icu API with zero extra DB costs. Future-proofed for Garmin, Strava, COROS, and TrainingPeaks.
🧬 13 Physiological Analytical Engines:
Threshold Engine: Multi-model synthesis (Daniels, Norwegian Method, Casado, Seiler, Critical Speed, LT1, LT2).
TID Engine: Polarized (80/20), Pyramidal, and Threshold distribution classification.
Performance & PMC Engine: CTL (Fitness), ATL (Fatigue), TSB (Form), VDOT, Running Economy.
Prediction Engine: Riegel exponential race time predictor (1K to Marathon).
Recovery Engine: Readiness scoring (0-100), HRV balance, resting HR trends, deload recommendation.
Injury Risk Engine: ACWR (Acute:Chronic Workload Ratio 7d vs 28d EWMA), Monotony, Strain.
Workout Builder Engine: Structured step workout generator (Easy, Threshold Cruise Intervals, VO2max, Reps).
Nutrition Engine: Carb Loading (g/kg), hydration (mL/hr), sodium, intra-race gel schedule.
Race Planner Engine: Negative split strategy planner & KM split table.
Report Engine: Full GitHub Markdown performance report generator.
⚡ 70+ Dedicated MCP Tools: Native support for Stdio (Claude Desktop/Cursor/Windsurf) and Express SSE (Claude.ai / Remote Connectors).
💸 100% Free Hosting Deployment: Zero database dependencies required. Ready for Railway, Render, Fly.io, or Docker.
Related MCP server: fitness-agent-mcp
git clone https://github.com/your-username/tracklab-mcp.git
cd tracklab-mcp
npm installCopy .env.example to .env:
INTERVALS_API_KEY=your_intervals_api_key
INTERVALS_ATHLETE_ID=i00000
MCP_TRANSPORT=stdionpm run build
npm startSee detailed configuration instructions in docs/CLIENT_SETUP.md.
Claude Desktop: Connect via STDIO transport in claude_desktop_config.json.
Cursor / Windsurf: Add node dist/index.js under MCP Settings.
Claude.ai: Deploy via Docker/Railway and point to https://your-app.up.railway.app/sse.
Run Vitest unit tests for engines:
npm testTypecheck TypeScript strict build:
npm run typecheckMIT License.
Analyzes cardiac decoupling (% aerobic drift) for a specific activity stream.
Inputs: activityId (required).
Executes multi-literature threshold analysis (Daniels, Norwegian Double Threshold, Casado, Seiler 3-Zone) with confidence scoring.
Inputs: no named parameters.
Generates structured step workout (Threshold Cruise Intervals or VO2max Intervals).
Inputs: type (required), repetitions.
Calculates Acute:Chronic Workload Ratio (7d vs 28d EWMA), Monotony, and Strain.
Inputs: no named parameters.
Calculates Carb Loading (g/kg), fluid (mL/hr), sodium, and gel schedule.
Inputs: targetRaceHours (required).
Calculates 3-zone Training Intensity Distribution (Polarized 80/20 vs Pyramidal vs Threshold).
Inputs: days.
Calculates Jack Daniels VDOT rating from a race performance.
Inputs: timeSeconds (required), distanceMeters (required).
Exports weekly performance & PMC summary report in GitHub Markdown.
Inputs: no named parameters.
Generates negative split kilometer pacing chart and fuel strategy.
Inputs: distanceKm (required), targetTimeSeconds (required).
Fetches athlete profile, LTHR, LT Pace, VDOT, and body metrics.
Inputs: no named parameters.
Calculates Banister EWMA PMC (CTL Fitness, ATL Fatigue, TSB Form).
Inputs: no named parameters.
Calculates athlete 5-zone HR breakdown based on LTHR & Max HR.
Inputs: no named parameters.
Fetches the most recent workout session and summary.
Inputs: no named parameters.
Lists all scientific paper & textbook references (Daniels, Seiler, Norwegian Method, Casado, Riegel, Banister, ACWR) registered in TrackLab.
Inputs: no named parameters.
Calculates athlete pace zones (Z1 Easy to Z5 Repetition).
Inputs: no named parameters.
Calculates 0-100 Training Readiness score and workout recommendation.
Inputs: no named parameters.
Lists recent activities with pace, HR, distance, and training load.
Inputs: limit.
Predicts race finish time using Riegel & Daniels models with validity checking.
Inputs: knownTimeSeconds (required), knownDistanceMeters (required), targetDistanceMeters (required).
Pings provider API endpoint and reports response latency.
Inputs: no named parameters.
Checks active endurance provider name, capabilities (HR, Power, Streams, HRV, Sleep), and health status.
Inputs: no named parameters.
Unified Status Dashboard summarizing PMC, TID, ACWR, Readiness, and Provider Status.
Inputs: no named parameters.
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
Scored across 21 tools
Most tools target distinct analytical functions (e.g., calculate_vdot vs predict_race), but some pairs like get_latest_activity vs list_recent_activities and get_fitness_fatigue_form vs get_training_readiness could be confused. Descriptions generally clarify the differences.
The majority follow a verb_noun pattern (get_*, calculate_*, analyze_*), but several deviate: provider_status, provider_ping, and training_dashboard are noun-led. This mixed convention is readable but not fully predictable.
With 21 tools, the server is at the higher end of the reasonable range for a comprehensive endurance analytics platform. Each tool has a specific purpose, and the count aligns with the 'Ultimate Edition' scope, though it feels slightly heavy.
The domain of training analytics is well covered with load metrics, threshold analysis, and race planning, but there are notable gaps: no tool to fetch a single activity's detailed stream (only analyze_activity_drift references streams), and no explicit activity detail view. This could force workarounds.
Adaptive running coach MCP server — training data, plans, and recovery for AI assistants.
AI coach for Garmin: builds training plans and structured workouts, synced straight to your watch.
Running coach: your AI reads your runs and recovery, and writes sessions to your watch or phone.
Your Garmin, Oura, WHOOP, Hevy and 35 more fitness apps, queryable from your own AI.