convez.ai
SkillIntegração & APIsAvançado
jacKlinc

Strava Claude - Treinador de Corrida com IA

jacKlinc/strava-claude

Analise corridas e treinos do Strava com coaching automático do Claude

Ver no GitHub
InicialPluginScore · de 0 a 100

Skill que integra dados de atividades Strava com Claude para análise automática de treinos. Funciona via Lambda AWS com OAuth transparente, oferecendo coaching personalizado com métricas de desempenho, deriva cardíaca e recomendações de treino.

Instalação

$npx convez-ai jacKlinc/strava-claude

1 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

Estrelas
0
Forks
0
Acompanhando
0
Saúde
28%
Último commit
há 3 meses
Versão
—
Issues abertas
1
Contribuidores
1

Casos de uso

  • Analisar métricas de corridas (pacing, elevação, frequência cardíaca)
  • Detectar padrões de desempenho e deriva cardíaca em treinos
  • Receber recomendações personalizadas de treinamento baseadas em dados

Veredito

Integração bem arquitetada entre Strava e Claude com deploy simplificado via CDK. Requer configuração AWS prévia, mas oferece valor real para corredores que querem análise inteligente de treinos sem gerenciar tokens manualmente.

README

A tradução para o português deste README ainda está na fila.

strava-claude

A personal running coach powered by Claude, backed by an AWS Lambda proxy that handles Strava OAuth so you never have to. Works in Claude Code (via a skill) and Claude.ai web/mobile (via MCP).

[!NOTE] AI Transparency Disclosure: This project utilizes AI coding assistants to generate boilerplates, optimize benchmarks, and refine documentation. All critical logic and performance calculations are human-reviewed and verified.

Architecture

Claude Code / Claude.ai  →  Lambda Function URL  →  Secrets Manager (OAuth creds)
                                                 →  Strava API

The Lambda refreshes your Strava access token on every invocation — no token rotation to manage.

Prerequisites

  • AWS CLI configured (aws sts get-caller-identity should work)
  • CDK bootstrapped in your target account/region (cdk bootstrap)
  • A Strava API app — create one at strava.com/settings/api
  • Go 1.21+ (for building the Lambda)

Setup

1. Configure your .env

Copy the example and fill in your values:

cp .env.example .env
VariableDescription
CLIENT_IDStrava app client ID
CLIENT_SECRETStrava app client secret
REFRESH_TOKENLong-lived Strava refresh token with activity:read_all scope — see Getting a refresh token
SKILL_SECRETA secret string you choose — used to protect your Lambda URL
BASE_URLSet this after the first deploy (copied from CDK output)
TRAINING_GOALAppears in the coaching persona, e.g. training for a marathon

2. Create the AWS secret

The CDK stack reads credentials from Secrets Manager. Create it from your .env:

make create-secret

Verify it looks right:

make check-secret

3. Deploy the Lambda

make deploy

This compiles the Go Lambda for linux/arm64, zips it, and runs cdk deploy. Copy the URL from the output and add it to your .env as BASE_URL:

Outputs:
StravaSkillStack.StravaSkillStackApiURL = https://xxxx.execute-api.region.amazonaws.com

4. Install the Claude Code skill

make install-skill

This substitutes BASE_URL, SKILL_SECRET, and TRAINING_GOAL into the skill template and writes it to ~/.claude/skills/strava-analyzer/SKILL.md.

5. Connect Claude.ai (web/mobile) via MCP

In Claude.ai → Settings → Integrations → Add MCP Server, enter:

  • URL: https://<BASE_URL>/mcp?key=<SKILL_SECRET>

No other fields needed. Claude.ai's connector will discover the four tools automatically (list_activities, get_activity, get_streams, get_laps).

To give the web/mobile Claude the same coaching persona, paste the contents of skill/SKILL.md (after running make install-skill so the variables are filled in) as a Project instruction in Claude.ai.

Usage

Claude Code — just ask naturally:

analyse my last run
how was my workout today
check my cardiac drift from yesterday's long run

Claude.ai — same phrases work once the MCP server is connected and the Project instruction is set.

Claude fetches your activity, streams, and laps, then delivers a structured coaching breakdown: pacing strategy, cardiac drift (aerobic decoupling %), elevation impact, and next-session recommendations.

Getting a Strava refresh token

The Lambda uses a long-lived refresh_token to mint fresh short-lived tokens on every call. You need a one-time OAuth exchange to get it.

Step 1 — Authorise your app in the browser. Build the authorization URL from your Strava app settings page, making sure scope=activity:read_all is included. After approving, Strava redirects to something like http://localhost/?code=XXXX — copy the code.

The code is single-use and expires in minutes. Exchange it immediately — do not paste it into .env or Secrets Manager.

Step 2 — Exchange the code and write to Secrets Manager in one shot (avoids copy-paste truncation):

RESPONSE=$(curl -s -X POST https://www.strava.com/oauth/token \
  -d client_id=YOUR_CLIENT_ID \
  -d client_secret=YOUR_CLIENT_SECRET \
  -d code=PASTE_CODE_HERE \
  -d grant_type=authorization_code) && \
REFRESH=$(echo "$RESPONSE" | grep -o '"refresh_token":"[^"]*"' | cut -d'"' -f4) && \
echo "refresh_token length: ${#REFRESH}" && \
aws secretsmanager put-secret-value \
  --secret-id StravaSkillStack/strava/oauth \
  --secret-string "{\"client_id\":\"YOUR_CLIENT_ID\",\"client_secret\":\"YOUR_CLIENT_SECRET\",\"refresh_token\":\"$REFRESH\",\"skill_auth_key\":\"YOUR_SKILL_SECRET\"}"

The command prints the token length — it should be 40. If shorter, print $RESPONSE to see the error.

Step 3 — Verify the token has the right scope:

curl -s -X POST https://www.strava.com/oauth/token \
  -d client_id=YOUR_CLIENT_ID \
  -d client_secret=YOUR_CLIENT_SECRET \
  -d refresh_token=$(aws secretsmanager get-secret-value \
      --secret-id StravaSkillStack/strava/oauth \
      --query SecretString --output text | grep -o '"refresh_token":"[^"]*"' | cut -d'"' -f4) \
  -d grant_type=refresh_token | grep -o '"scope":"[^"]*"'

Should return "scope":"activity:read_all,read". If it only shows "scope":"read", the OAuth flow was done without activity:read_all — repeat from Step 1.

Also update REFRESH_TOKEN in your .env so make create-secret stays in sync.

Integration tests

Run the full suite against the live endpoint:

make test

Covers REST endpoints (activities, detail, streams, laps, auth) and all MCP methods (initialize, tools/list, each tool, error cases, notifications).

Makefile targets

TargetDescription
make create-secretCreate the Secrets Manager secret from .env values
make check-secretPrint the current secret value
make build-lambdaCross-compile Go for linux/arm64 and zip
make deployBuild + cdk deploy
make install-skillRender skill template and copy to ~/.claude/skills/strava-analyzer/
make testRun integration tests against the live endpoint
make cleanRemove build artifacts
Ver o original no GitHub

Bom para

Relacionados