AI fitness coaching through any MCP-compatible AI assistant. Plan training, log workouts, track benchmarks, manage goals, and get coaching insights — all through natural conversation.
Website · Integrations Guide · How It Works · Methodology
MCP Server URL: https://api.pelaris.io/mcp
Settings → Apps → Add → enter the MCP Server URL above
Settings → Connectors → Add Custom → enter the MCP Server URL above → Advanced Settings → Client ID: pelaris-claude
Connect to https://api.pelaris.io/mcp — supports OAuth 2.0 with PKCE and Dynamic Client Registration.
Related MCP server: intervals-icu-mcp
Tool | Description |
| View your training context, active programs, and recent sessions |
| View current program with phase, weekly structure, and session details |
| View a specific session's exercises, sets, targets, and feedback |
| Resolve exercise names to canonical exercise IDs before planning or logging |
| View benchmark values, progress history, and trends |
| View body composition data and measurement trends |
| Search curated training articles and resources |
| Get data-driven coaching insights based on your training |
| Check profile setup completion status |
| View weekly training summary and coaching focus |
| Check the status and progress of your training programs (running, paused, replaced or removed) |
| List your training goals with completion status and linked benchmarks |
Tool | Description |
| Complete onboarding intake and persist your training profile |
| Get the link that starts your training program in Pelaris (programs are built in Pelaris) |
| Create a planned workout with exercises and targets |
| Log a completed workout or mark a planned session as done |
| Log a full completed session with exercises, sets, and feedback |
| Get alternative exercise suggestions |
| Adjust session volume, intensity, or schedule |
| Update the details of an existing planned session |
| Delete a single planned session |
| Delete multiple planned sessions |
| Record an injury with body part, severity, and notes |
| Update equipment, availability, and preferences |
| Submit coaching quality feedback |
| Record a benchmark value with history tracking |
| Log daily readiness, soreness, and sleep quality |
| Create, update, or complete training goals |
| Set a program aside: removes one you have not started, and points you to Pause, Stop and Remove in the app for the rest |
Once connected, talk to your AI assistant naturally. It maps your intent to the tools above:
"What does my training look like today?" → get_training_overview / get_active_program
"Log my 5k run, felt easy, RPE 4." → log_workout
"Create an upper body session for tomorrow." → create_planned_session
"How are my benchmarks trending?" → get_benchmarks
"Swap the barbell bench for a dumbbell variation." → swap_exercise
"My right shoulder's been sore." → record_injury
"I want a training program." → generate_program (returns your Pelaris link, where your program is built)
(Illustrative prompts only; no personal data is shown. Responses are grounded in your own training data and PII-scrubbed.)
OAuth 2.0 with PKCE. The server supports:
Pre-registered clients for ChatGPT and Claude
Dynamic Client Registration for all other MCP clients
Strength · Running · Swimming · Cycling · Triathlon · CrossFit · General Fitness
Pelaris implements 28 science-based training methodologies. Learn more about our methodology.
Pseudonymous user IDs (Firebase UIDs are never exposed)
PII scrubbing on all responses
Granular OAuth scopes
Users can disconnect anytime
Privacy Policy · Terms of Service
npm run build, then npm test. Every test file is listed in scripts/run_tests.mjs, and Firestore points at a dead loopback port unless an emulator is set, so a test never reaches a real project. CI runs both before each deploy.
Local runs use scripts/run_local_gateway.sh. It refuses to start unless Firestore, the OAuth function and the core bridge are all on loopback, the port is not an emulator port, and the secrets are the local test values.
End to end, locally: with the Pelaris local stack up (emulators, functions and web, sends off), node tools/connector_e2e/run_connector_e2e.mjs --reseed --start-gateway --transcript <file> starts this gateway under wayfinder's egress guard and drives it with a real MCP client as seeded people (newcomer, program holder, unstarted, started and two-program athletes), then checks the replies, the emulator's data and the gateway log. Local only: it refuses anything that is not loopback and mints test tokens with the runner's published dummy secret. The header of the script lists the people and the flags.
The contracts shared with the Pelaris backend live in test_fixtures/ (redirect policy, token ownership, return predicates). Each is pinned by its SHA-256 on both sides.
This server cannot be deployed
Pace is a remote MCP server that exposes wearable and fitness data to Claude via the Model Context Protocol. It connects to Garmin, Oura, Whoop, Polar, Fitbit and 20+ devices and provides 15 tools for querying sleep, activity, recovery, and training data. Hosted on Google Cloud Run, OAuth 2.1 authentication, Streamable HTTP transport. Instructions: First you need to create an account at: https://pacetraining.co and connect your wearables. After that you can connect the remote Server via Custom Connector in Claude and OAuth 2.1 Flow startet.
Adaptive running coach MCP server — training data, plans, and recovery for AI assistants.
tri-train is a training platform for triathletes that syncs activities from Strava and Garmin and exposes them to Claude through a remote MCP server. Ask Claude how your last block went, why your run pace dropped, or whether you are fresh enough for Sunday's race, and it answers from your real data: activities with laps and grade-adjusted pace, weekly load and the performance management chart (CTL/ATL/TSB), sleep, HRV and resting heart rate, and your race calendar with predictions. Claude can also act: create and adjust planned workouts with structured intervals, push them to your Garmin watch, add races and write notes. Every change is journaled and can be undone. Read and write access are separate OAuth scopes, and the server enforces them on every call. Built and hosted in Switzerland by a triathlete; data stays on our own servers (Swiss revDSG / GDPR).
Your Garmin, Oura, WHOOP, Hevy and 35 more fitness apps, queryable from your own AI.
