Pace-AI

AI running coach powered by MCP — connects Claude to your Strava, Garmin, Withings, and Notion data for evidence-based coaching, structured workout delivery to your watch, and a web-based coaching UI.

Five MCP servers in one monorepo, plus a Flask web UI that orchestrates multi-prompt coaching sessions. Claude pulls training data from Strava, wellness from Garmin, body composition from Withings, diary entries from Notion, reasons using a 3,113-claim sports-science evidence base, and pushes structured workouts to your Garmin watch.

STATUS Agent — Parallel Execution

Web UI

The Flask UI (localhost:5050) provides a coaching dashboard with 6 focused agents. All agent outputs use structured JSON rendered into styled section cards with RAG status indicators (green/amber/red). Professional dark-themed UI with Inter font, muted palette, and 1120px layout.

Agent	Trigger	Context	Purpose
STATUS	Status button	Profile, activities, wellness, body comp, Garmin calendar, diary, coaching log	Full training status assessment — 3 parallel Claude calls (training, recovery, injury) + sequential readiness pass. Body composition and schedule rendered directly from data.
PLAN	Plan button + date range	Status snapshot, research evidence, athlete facts, coaching context	Generate evidence-based training plan as structured JSON (report sections + sessions table)
CHAT	Text input	Profile summary, facts, coaching context, pending plan	Conversational Q&A and plan tweaks
EXERCISE	Schedule button	Plan JSON only	Add structured exercises to strength/mobility sessions
NUTRITION	Nutrition button + mode	Profile, status snapshot, research claims, athlete facts (nutrition category)	Sports nutrition advice in 3 modes: general, plan-paired, race fueling
END SESSION	End Session button	Full conversation history, coaching context	Summarise session, update coaching memory

python -m ui.app    # opens http://localhost:5050

Workflow: Sync All → Status → Plan → tweak via chat → Schedule → End Session

Quick Start

1. Create API Apps

Strava: strava.com/settings/api — callback domain localhost
Withings: developer.withings.com — create an app
Notion: notion.so/my-integrations — internal integration with read access to your diary database
Garmin: No API app needed — uses Garmin Connect SSO login

2. Install

git clone https://github.com/aldred-coetzee/Pace-AI.git
cd Pace-AI
pip install -e ./strava-mcp -e ./pace-ai -e ./garmin-mcp -e ./withings-mcp -e ./notion-mcp

3. Configure

cp strava-mcp/.env.example strava-mcp/.env
# Edit: STRAVA_CLIENT_ID, STRAVA_CLIENT_SECRET

cp garmin-mcp/.env.example garmin-mcp/.env
# Edit: GARMIN_EMAIL, GARMIN_PASSWORD

cp withings-mcp/.env.example withings-mcp/.env
# Edit: WITHINGS_CLIENT_ID, WITHINGS_CLIENT_SECRET

cp notion-mcp/.env.example notion-mcp/.env
# Edit: NOTION_TOKEN, NOTION_DIARY_DATABASE_ID

Garmin requires a one-time login (handles MFA):

garmin-mcp-login

4. Start Servers

strava-mcp &       # localhost:8001
pace-ai &          # localhost:8002
garmin-mcp &       # localhost:8003
withings-mcp &     # localhost:8004
notion-mcp &       # localhost:8005

5. Use

Web UI (recommended):

python -m ui.app   # localhost:5050

Claude Desktop / Claude Code — add to MCP config:

{
  "mcpServers": {
    "strava-mcp": { "type": "streamableHttp", "url": "http://127.0.0.1:8001/mcp" },
    "pace-ai": { "type": "streamableHttp", "url": "http://127.0.0.1:8002/mcp" },
    "garmin-mcp": { "type": "streamableHttp", "url": "http://127.0.0.1:8003/mcp" },
    "withings-mcp": { "type": "streamableHttp", "url": "http://127.0.0.1:8004/mcp" },
    "notion-mcp": { "type": "streamableHttp", "url": "http://127.0.0.1:8005/mcp" }
  }
}

MCP Servers

strava-mcp — Data Access (12 tools, 2 resources)

Tool	Description
`authenticate`	Trigger Strava OAuth flow
`get_athlete`	Athlete profile
`get_recent_activities`	Recent activities with pace, HR, elevation
`get_activity`	Full activity detail (splits, laps)
`get_activity_streams`	Time-series data (HR, GPS, cadence, altitude)
`get_athlete_stats`	Year-to-date and all-time stats
`get_athlete_zones`	Heart rate and power zone definitions
`get_best_efforts`	Personal bests for standard distances
`get_weekly_summary`	Rolling weekly aggregates
`get_shoe_mileage`	Shoe mileage with retirement warnings
`search_activities`	Search/filter activities by type, distance, name
`get_segment_analysis`	Repeated segment comparison over time

pace-ai — Coaching Intelligence (43 tools, 4 prompts, 3 resources)

Analysis (8 tools): ACWR (weekly + daily EWMA), VDOT race prediction, Daniels training zones, Karvonen HR zones, cardiac decoupling, fitness trend assessment, race readiness scoring.

Run Analysis (5 tools): Structured single-run analysis, workout type auto-classification, intensity distribution / 80-20 check, data anomaly detection.

Sync (7 tools): Incremental sync from Strava, Garmin wellness, Garmin workouts, Withings, Notion. Sync status tracking. One-call sync_all.

History (7 tools): Weekly distances, recent activities (local), wellness snapshots, diary entries, race history, personal bests.

Coaching Memory (8 tools): Append-only coaching log, coaching context (read/rewrite), athlete facts (injury, preference, goal, training response), coaching log search.

Profile (3 tools): Auto-generate profile from synced data, get/update manual profile fields.

Goals (4 tools): CRUD for race goals (type, target time, date, notes).

Environment (2 tools): Heat and altitude pace adjustments.

Evidence (1 tool): Query 3,113 peer-reviewed claims across 54 research categories.

Prompt	Description
`weekly_plan`	Weekly training plan with research evidence
`run_analysis`	Post-run coaching analysis
`race_readiness`	Race readiness assessment
`injury_risk`	Injury risk from load patterns

garmin-mcp — Workout Delivery + Wellness (17 tools, 1 resource)

Workouts (10 tools): Create, list, get, delete, schedule, unschedule workouts. Combined create-and-schedule. Calendar listing. Supports easy_run, run_walk, tempo, intervals, strides, strength, mobility, yoga, cardio, hiit, walking, custom workout types with HR zone targeting. SPORT_TYPE_MAP and resolve_sport_type() map session types to FR245-compatible Garmin sport types.

Wellness (7 tools): Body battery, sleep score, HRV, training readiness, stress, resting HR, combined wellness snapshot.

withings-mcp — Body Composition (5 tools, 1 resource)

Tool	Description
`authenticate`	Check Withings connection
`get_measurements`	Weight, body fat %, muscle mass, bone mass, body water
`get_latest_weight`	Most recent weight measurement
`get_blood_pressure`	Systolic, diastolic, heart rate
`get_body_composition_trend`	Weekly averages over time

notion-mcp — Running Diary (1 tool, 1 resource)

Tool	Description
`get_diary_entries`	Sync and return diary entries (stress 1-5, niggles, notes)

Research Evidence

The coaching system is grounded in a database of 3,113 claims extracted from 520 peer-reviewed papers across 54 research categories. Claims are scored by population relevance and injected into coaching prompts via RAG.

Categories always queried for weekly plans:

Category	Claims	Coverage
strength_training_runners	64	Running economy, frequency, exercise selection
foam_rolling_mobility	47	ROM effects, warm-up vs recovery, protocols
warmup_cooldown	72	Static vs dynamic stretching, optimal warm-up
recovery_modalities	54	Compression, cold water, massage, foam rolling
concurrent_training	59	Strength + running interference, programming
overtraining_recovery	60	Load monitoring, monotony, session RPE
sleep_recovery	60	Sleep loss effects, performance impact
training_load_acwr	70	ACWR thresholds, injury risk, load management
injury_prevention_general	57	Gait retraining, strength for injury reduction
easy_recovery_running	49	Aerobic base, fat oxidation, mitochondrial

Conditionally added: tendon_health, injury_lower_leg, injury_knee, injury_stress_fracture, return_to_running, detraining, marathon/half/5k training, masters_running.

Configuration

Variable	Default	Description
`STRAVA_CLIENT_ID`	(required)	From Strava API settings
`STRAVA_CLIENT_SECRET`	(required)	From Strava API settings
`STRAVA_MCP_PORT`	`8001`	strava-mcp HTTP port
`PACE_AI_PORT`	`8002`	pace-ai HTTP port
`GARMIN_EMAIL`	(required)	Garmin Connect login
`GARMIN_PASSWORD`	(required)	Garmin Connect password
`GARMIN_MCP_PORT`	`8003`	garmin-mcp HTTP port
`GARTH_HOME`	`~/.garth`	Garth session token directory
`WITHINGS_CLIENT_ID`	(required)	From Withings developer portal
`WITHINGS_CLIENT_SECRET`	(required)	From Withings developer portal
`WITHINGS_MCP_PORT`	`8004`	withings-mcp HTTP port
`NOTION_TOKEN`	(required)	Notion internal integration secret
`NOTION_DIARY_DATABASE_ID`	(required)	Notion database ID for diary
`NOTION_MCP_PORT`	`8005`	notion-mcp HTTP port

Development

# Install in dev mode
pip install -e ./strava-mcp[dev] -e ./pace-ai[dev] -e ./garmin-mcp[dev] -e ./withings-mcp[dev] -e ./notion-mcp[dev]

# Run tests (all servers)
cd strava-mcp && python -m pytest tests/ && cd ..
cd pace-ai && python -m pytest tests/ && cd ..
cd garmin-mcp && python -m pytest tests/ && cd ..
cd withings-mcp && python -m pytest tests/ && cd ..
cd notion-mcp && python -m pytest tests/ && cd ..

# Lint
ruff check strava-mcp/ pace-ai/ garmin-mcp/ withings-mcp/ notion-mcp/ ui/
ruff format --check strava-mcp/ pace-ai/ garmin-mcp/ withings-mcp/ notion-mcp/ ui/

~628 test functions (more with parametrize) across unit, integration, and e2e suites. E2e tests boot each server as a subprocess and verify HTTP response.

See individual server READMEs for server-specific details: strava-mcp, pace-ai, garmin-mcp, withings-mcp, notion-mcp.

License

MIT

Navigation Menu

Repository navigation

About

Repository files navigation

Pace-AI

STATUS Agent — Parallel Execution

Web UI

Quick Start

1. Create API Apps

2. Install

3. Configure

4. Start Servers

5. Use

MCP Servers

strava-mcp — Data Access (12 tools, 2 resources)

pace-ai — Coaching Intelligence (43 tools, 4 prompts, 3 resources)

garmin-mcp — Workout Delivery + Wellness (17 tools, 1 resource)

withings-mcp — Body Composition (5 tools, 1 resource)

notion-mcp — Running Diary (1 tool, 1 resource)

Research Evidence

Configuration

Development

License

About

Releases

Packages

Contributors

Languages

Name		Name	Last commit message	Last commit date
docs		docs
garmin-mcp		garmin-mcp
notion-mcp		notion-mcp
pace-ai		pace-ai
research		research
research_archive		research_archive
strava-mcp		strava-mcp
tests		tests
ui		ui
withings-mcp		withings-mcp
.gitignore		.gitignore
CLAUDE.md		CLAUDE.md
LICENSE		LICENSE
README.md		README.md

Latest commit

History

Folders and files

Repository files navigation

Pace-AI

STATUS Agent — Parallel Execution

Web UI

Quick Start

1. Create API Apps

2. Install

3. Configure

4. Start Servers

5. Use

MCP Servers

strava-mcp — Data Access (12 tools, 2 resources)

pace-ai — Coaching Intelligence (43 tools, 4 prompts, 3 resources)

garmin-mcp — Workout Delivery + Wellness (17 tools, 1 resource)

withings-mcp — Body Composition (5 tools, 1 resource)

notion-mcp — Running Diary (1 tool, 1 resource)

Research Evidence

Configuration

Development

License

About

Resources

Stars

Watchers

Forks

Releases

Packages

Contributors

Languages