strava-claude

A personal MCP (Model Context Protocol) server that gives Claude access to your Strava training data. Built for personal use with Claude.ai and the Claude mobile app.

---

What it does

Exposes a single tool — get_strava_activities — that Claude can call to fetch your recent Strava activities directly from the Strava API. Each activity includes:

• Date, sport type, name, and description
• Distance (miles), moving time, elapsed time, elevation
• Average and max heart rate (where recorded)
• Average pace per mile (for runs)
• Lap-by-lap breakdown

When added to a Claude project with the instruction to call this tool at the start of each conversation, Claude automatically loads your latest data without you having to upload or paste anything.

What it can't do

• Write to Strava — read-only, no creating or modifying activities
• Fetch more than 20 activities by default (configurable via num_activities parameter)
• Authenticate via OAuth — uses no auth; intended for personal/private deployment only
• Run without a server — requires a VPS or other always-on host to work with Claude.ai and the mobile app (a local stdio version for the Claude desktop app is also included)

Project structure

strava_sync.py              # Syncs activities to activities.txt (used by local/desktop version)
strava_mcp_server.js        # Local MCP server (stdio) — for Claude desktop app only
strava_mcp_server_http.js   # Remote MCP server (HTTP) — for Claude.ai and mobile app
deploy.sh                   # Deploys HTTP server to VPS
.env.example                # Required environment variables

Setup

Requirements

• Node.js 18+
• A Strava API app (create one here)
• For remote use: a VPS with nginx and a domain or sslip.io address

Environment variables

Copy .env.example to .env and fill in your values:

STRAVA_CLIENT_ID=your_client_id
STRAVA_CLIENT_SECRET=your_client_secret
STRAVA_REFRESH_TOKEN=your_refresh_token

Run locally (Claude desktop app)

npm install

Add to ~/Library/Application Support/Claude/claude_desktop_config.json:

{
  "mcpServers": {
    "strava": {
      "command": "node",
      "args": ["/path/to/strava_mcp_server.js"]
    }
  }
}

Deploy remotely (Claude.ai + mobile)

./deploy.sh

Requires nginx and certbot configured on the VPS. The MCP endpoint will be available at https://yourdomain.com/mcp.

Add as a custom connector in Claude.ai under Settings → Connectors → Add custom connector.

Claude project instructions

Add this to your Claude project instructions so Claude fetches data automatically at the start of each conversation:

At the start of every new conversation, call the get_strava_activities tool to fetch my latest training data. Do this before responding to anything else. Then provide a brief training summary.