robo-trainer-mcp

MCP server that turns Claude into a full-stack personal trainer. Nutrition coaching, strength programming, powerlifting analytics, strongman tracking, recovery monitoring, and mental performance: all from a single local server, nothing leaving your machine.

The problem it solves

Every fitness app works in isolation. Your food tracker knows your macros. Your training log knows your lifts. Your sleep tracker knows your HRV. None of them talk to each other, and none of them can tell you why you're stuck.

robo-trainer-mcp connects everything. Claude can pull training volume, sleep trends, nutrition, and mood at once: then reason across them like a real coach.

Demos

Diagnosing a plateau

You: "My squat hasn't moved in 5 weeks and I feel exhausted. What's wrong?"

Claude pulls:
  training_get_volume_trend("squat", weeks=6)
  recovery_get_sleep_trend(days=42)
  nutrition_get_weekly_macros()
  mental_get_mood_trend(days=42)

  Squat volume: 12,400 → 15,300 kg  +23%  (no deload)
  Sleep:        8.1h → 6.3h          ▼ 22%
  Calories:     2,150 kcal vs 2,600 goal   deficit −450/day
  Protein:      168 g vs 210 g goal        gap −42 g/day

  Diagnosis: Classic overreaching. Four recovery systems degraded
  simultaneously. Deload this week, calories to maintenance,
  8+ hours sleep. Expect a squat PR within 3 weeks.

Powerlifting meet prep

You: "I squat 200, bench 130, deadlift 240 at 93 kg bodyweight."

  Wilks 2020:  358.4  (Advanced)
  DOTS:        347.1  (Advanced)

  Weakest lift: Bench: 3.2% below elite average proportion.
  Recommendation: hypertrophy + technique block on bench.

  Attempt suggestions:
    Squat    180 / 195 / 207.5
    Bench    117.5 / 127.5 / 135
    Dead     215 / 235 / 250

Readiness check

You: "I slept 6 hours, quality 5/10, soreness 7/10, stress 7/10."

  recovery_calculate_readiness(6.0, 5, 7, 7)

  Readiness: 41 / 100
  Recommendation: light technique session or active recovery.
  Heavy singles today carry elevated injury risk.

Architecture

                 ┌──────────────────────────────────┐
                 │          Claude (LLM)             │
                 │   reasons across all 24 tools     │
                 └─────────────┬────────────────────┘
                               │  MCP Protocol
                 ┌─────────────▼────────────────────┐
                 │        robo-trainer-mcp           │
                 │         (FastMCP server)          │
                 └────────┬─────────────┬────────────┘
                          │             │
             ┌────────────▼──┐  ┌─────▼─────────────────┐
             │   SQLite DB   │  │  Open Food Facts API  │
             │ ~/.robo-      │  │  (food search only)   │
             │  trainer/     │  └──────────────────────┘
             │  data.db      │
             │               │
             │  food_log     │
             │  training_log │
             │  sleep_log    │
             │  mood_log     │
             └───────────────┘

All personal data stays in ~/.robo-trainer/data.db. No account required.

Tools: 27 total

Nutrition

Strength Training

Powerlifting

Strongman

Recovery

Mental Performance

Quick start

Not yet published to PyPI. Install from source or use the Docker image below.

git clone https://github.com/gerardrecinto/robo-trainer-mcp
cd robo-trainer-mcp
pip install -e .
robo-trainer-mcp

Docker

docker pull ghcr.io/gerardrecinto/robo-trainer-mcp:latest

Add to Claude Desktop (~/Library/Application Support/Claude/claude_desktop_config.json):

{
  "mcpServers": {
    "trainer": {
      "command": "robo-trainer-mcp"
    }
  }
}

Resilient module loading

Each tool module (nutrition, training, recovery, etc.) loads independently. If a module fails to import, the server logs the error and continues loading the rest. Version is logged at startup. This means the server comes up cleanly even if one module has a dependency issue.

Tests

pip install -e ".[dev]"
pytest tests/ -v
# 16 passed

Product Direction

This can grow into a local-first AI coaching product: open-source MCP server as the wedge, Pro dashboards for athletes, and Coach Mode for trainers managing clients.

See docs/go-to-market.md.

License

MIT