Intervals.icu MCP Server com Stryd
methylone/Intervals-MCP-Server-with-STRYD
PMC dual e análise de Stryd para corredores no Claude
Ver no GitHubServidor MCP local que expõe dados de treinamento do Intervals.icu para clientes IA (Claude, etc.), com cálculo server-side de PMC, desacoplamento cardíaco e análise de carga musculoesquelética via Stryd. Designed para corredores que usam power meters, oferecendo dual PMC (carga metabólica e musculoesquelética) sem deixar a matemática para a IA adivinhar.
Instalação
npx convez-ai methylone/Intervals-MCP-Server-with-STRYD1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
Casos de uso
- Análise de treinamento com power meter Stryd em ambiente IA
- Monitoramento de carga musculoesquelética vs. metabólica em ultras
- Integração de dados Intervals.icu em workflows com Claude Desktop
- Cálculo determinístico de PMC, decoupling cardíaco e tendências de impacto
Veredito

Servidor especializado e bem-arquitetado para atletas que combinam Stryd + Intervals.icu + Claude. O padrão de deixar matemática complexa server-side é sensato, mas exige setup técnico (variáveis de env, API keys, cache). Recomendado para corredores experientes em engenharia ou muito dedicados a análise de treinamento.
README
A tradução para o português deste README ainda está na fila.
English | 日本語
Intervals.icu MCP Server — for Stryd runners
An Intervals.icu MCP (Model Context Protocol) server built for Stryd runners — power-based running analysis (a dual Performance Management Chart with RSS, LBSS), designed so the LLM never does the math. The deterministic numbers (PMC values, cardiac decoupling, ramp rates) are computed server-side; the AI client (Claude Desktop, Codex, …) is left to interpret, not to guess.
It is designed to run locally, for a single athlete (you) — not as a hosted, multi-user service.
What it does
- Stryd extension (the reason this exists). A second Performance Management Chart
computed server-side from a lower-body load metric — LBSS (Lower Body Stress
Score) — via EMA, sitting next to Intervals.icu's built-in RSS-based PMC. You get a
dual PMC (musculoskeletal load and metabolic load), ILR (Impact Loading Rate)
trends, and weekly / phase-level summaries aimed at ultramarathon-style review. The
LBSS / ILR custom-field names are configurable (
LBSS_FIELD/ILR_FIELD, defaultStrydLBSSv2/StrydILR) so a recalibrated or renamed field needs no code change — see INSTALL.md and the field recipes.estimate_critical_impactreverse-estimates Stryd's Critical Impact from your Intervals streams and Critical Power (no Stryd API), so LBSS calibration stays self-contained. - Core (any Intervals.icu user). List and inspect activities, wellness, HRV trends, events / planned workouts (read + create / update / delete), athlete summaries, and stream-level analysis (splits, cardiac decoupling, grade-adjusted pace, custom power/HR zones).
The server provides data and math only. It does not decide how you should
train — that interpretation comes from a knowledge file you write and load into your
AI client. See training-knowledge-template/.
Quick start
Pick the install path that matches your client. Full steps and prerequisites: INSTALL.md.
1. MCPB bundle — Claude Desktop (easiest)
Download the .mcpb bundle from the
latest release,
double-click to install into Claude Desktop, and fill in the three fields it asks for
(Athlete ID, API key, timezone). Your API key is stored in the OS keychain, not in
a plaintext file.
2. npx — one-line config (Claude Desktop / Codex / any MCP client)
No clone, no build. Point your client at the published npm package:
{
"mcpServers": {
"intervals-stryd": {
"command": "npx",
"args": ["-y", "intervals-mcp-with-stryd"],
"env": {
"INTERVALS_ATHLETE_ID": "i0000000",
"INTERVALS_API_KEY": "your-api-key",
"ATHLETE_TIMEZONE": "Asia/Tokyo",
"CACHE_DIR": "/absolute/path/to/intervals-cache"
}
}
}
}
CACHE_DIR is optional but recommended under npx: without it the stream cache lands in
npx's volatile package cache. See INSTALL.md.
3. From source / Docker (development, HTTP mode)
git clone https://github.com/methylone/Intervals-MCP-Server-with-STRYD.git
cd Intervals-MCP-Server-with-STRYD
npm install
cp .env.example .env # then fill in your API key, athlete ID, timezone
npm run build
Then point your client at build/index.js over stdio, or run HTTP / Docker — see
INSTALL.md. New to this? Hand the repo URL to your AI client and ask it to
walk you through installation using the README and INSTALL.md.
Command-line use
The same tools are also available from a shell via the intervals-mcp CLI (no MCP
client, no LLM) — useful for automation, piping into jq, and quick checks. It returns
raw data only; it does not apply the methodology. See docs/CLI.md.
Documentation
- INSTALL.md — prerequisites and client setup (MCPB / npx / source)
- docs/CLI.md — running the tools from a shell
- ARCHITECTURE.md — code layout and how to extend it
- SECURITY.md — security & privacy model: what it connects to, what it reads/writes, where your key and cache live, HTTP mode
- ROADMAP.md — what's planned next
training-knowledge-template/— build your own analysis knowledge for your AI client
Security & privacy
It talks to one host (intervals.icu), writes only to your Intervals.icu calendar events, never logs your API key, and ships no telemetry. The HTTP transport has no application-layer authentication — run locally over stdio (or MCPB / npx) for personal use and never expose HTTP mode to the public internet. Full details — cache contents, key blast radius, build verification, uninstall — in SECURITY.md.
Contributing
Forks are welcome — take it and make it yours. Pull requests are not actively maintained, so please fork freely rather than expecting timely reviews.
License
AGPL-3.0-or-later. In short: you're free to use, modify, and run this, including commercially — but if you distribute it or run a modified version as a network service, you must release your source under the same license. It cannot be turned into a closed, proprietary product.