ai-cycling-coach
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., "@ai-cycling-coach review this week's training and replace tomorrow's workout with an easy ride"
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.
AI Cycling Coach
A small local stdio MCP server that gives compatible AI agents controlled access to one Intervals.icu account. Intervals.icu remains the source of truth and syncs planned workouts.
Codex setup (tested)
In Intervals.icu, open Settings → Developer Settings and generate a personal API key.
Run
scripts/setup.shand enter the key when prompted.Restart Codex.
Ask Codex to call
connection_statusbefore any other operation.
The key is stored only in .env, which is gitignored and set to owner-only
permissions. The setup script creates a local virtual environment and registers
scripts/run-server.sh as a local stdio MCP server named ai-cycling-coach.
Set PYTHON before running setup if you want to use a specific Python 3.11+
interpreter.
Related MCP server: TrainingPeaks MCP Server
Other MCP clients (untested)
The server uses standard MCP over stdio, so other MCP clients should be able to
launch the same scripts/run-server.sh command. These integrations have not
been tested yet.
Before registering another client, install the server and create its private environment file from the repository root:
python3.11 -m venv .venv
.venv/bin/python -m pip install -e .
read -r -s -p "Intervals.icu API key: " api_key
printf '\n'
umask 077
printf 'INTERVALS_API_KEY=%s\n' "$api_key" > .env
unset api_key
chmod u+x scripts/run-server.shUse a different Python 3.11+ executable if python3.11 is unavailable.
Claude Code
claude mcp add --scope user ai-cycling-coach -- \
/absolute/path/to/AiCyclingCoach/scripts/run-server.shOpenClaw
openclaw mcp add ai-cycling-coach \
--command /absolute/path/to/AiCyclingCoach/scripts/run-server.sh
openclaw mcp doctor ai-cycling-coach --probeDeepSeek
The DeepSeek API supports tool calling but does not launch local MCP servers by itself. Use DeepSeek through an MCP-capable host, such as OpenClaw, and register the server with that host.
Approval workflow
The agent may read training data automatically. Before writing, it must present
the complete plan outline and ask once for approval. Only then may it call
replace_training_plan with approved=true. Configure every MCP client to
require confirmation for this write tool: the server trusts the client's
approved value. The tool can replace only events created with the
ai-cycling-coach: ownership prefix.
First live verification
Run connection_status first and review the returned athlete identity. The
first calendar write should be a short visible test plan that you explicitly
approve.
Available Tools
7 toolsconnection_statusD
Inputs: no named parameters. Returns: object
get_activity_detailsD
Inputs: activity_id (required). Returns: object
get_athlete_profileD
Inputs: no named parameters. Returns: object
list_activitiesD
Inputs: newest (required), oldest (required). Returns: object
list_calendarD
Inputs: newest (required), oldest (required). Returns: object
list_wellnessD
Inputs: newest (required), oldest (required). Returns: object
replace_training_planC
Replace integration-owned events after approval of the complete plan.
Inputs: newest (required), oldest (required), approved (required), workouts (required), +1 more. Returns: object
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 7 tools
Each tool targets a distinct resource or level of detail: activities list vs. activity detail, wellness, calendar, athlete profile, connection status, and training plan replacement. There is no meaningful overlap between tool purposes.
The set mostly follows a get_/list_ + noun pattern, which is clear and predictable. Minor deviations are connection_status lacking a verb and the get_ vs. list_ split, but these do not seriously hurt readability.
Seven tools is well-scoped for an AI cycling coach integration, covering athlete data reads and the key plan replacement mutation without unnecessary redundancy.
The core read surfaces for activities, wellness, calendar, and athlete profile are covered, and replace_training_plan provides the key write operation. The main gap is the absence of explicit training-plan creation or approval tools, though the server may assume those happen externally.
Maintenance
Related MCP Connectors
Connect Intervals.icu to Claude and ChatGPT: read training, plan workouts. Training never stored.
Your endurance training data in your AI assistant: activities, recovery, plan, workout edits.
Connect Claude to your Intervals.icu watch data for fitness, workout review, and plan writing.
Your own cycling data for your AI assistant: rides, power, climbs and routes. Read-only.
Related MCP Servers
- FlicenseAqualityDmaintenanceEnables AI assistants to access and analyze intervals.icu training data including activities, fitness metrics (CTL/ATL/TSB), wellness stats, and calendar events. Supports natural language querying of athletic performance for training insights and workout planning.6-
- FlicenseNot gradedqualityDmaintenanceEnables AI assistants to interact with TrainingPeaks accounts to analyze training data, manage workouts, update zones, and more using plain English.3-
- AlicenseAqualityDmaintenanceProvides AI assistants with read-only access to an athlete's Intervals.icu training data, including activities, wellness metrics, zones, and planned events, for use with MCP clients like ChatGPT and Claude.15MIT
- AlicenseNot gradedqualityBmaintenanceEnables an AI agent to read Garmin sleep, readiness and activity data, build an athlete profile, check proposed sessions against code-enforced limits and real calendar availability, and then push approved structured workouts to a Garmin watch. Plans are proposed and stored under a plan ID, so only explicitly approved sessions are ever written.MIT