strava-mcp-agent

Your AI running coach that actually remembers. An MCP server with persistent memory that connects your Strava data to Claude — tracking your zones, pace trends, injuries, and goals across sessions so coaching advice stays current.

What it does

strava-mcp-agent gives Claude 18 tools — 13 for live Strava data and 5 for persistent memory that carries context across every conversation.

Memory & Context Tools

Strava Data Tools

Why memory matters

Without memory, Claude forgets everything between conversations:

• Session 1: "Your Z2 pace is 9:30/km"

• Session 2: "Your Z2 pace improved to 8:10/km"

• Session 3: "I recommend running at 9:00–9:30/km" ← wrong, forgot the improvement

With memory, fitness metrics are auto-computed from your Strava data and persist. Claude always knows your current zones, trends, and injuries.

Related MCP server: MCPacer

Quick Start (2 commands)

pip install strava-mcp-agent
strava-mcp-token

Step 1: Create a Strava API app (one time)

Go to strava.com/settings/api and fill in:

The callback domain must be localhost — this is what allows the setup wizard to receive the authorization code on your machine.

Click Create. On the next page, copy your Client ID (a number like 123456) and Client Secret (a long code like abc123def456...).

Step 2: Run the setup wizard

strava-mcp-token

It will:

1. Ask for your Client ID and Client Secret

2. Open your browser — click Authorize on the Strava page

3. Auto-detect your OS (macOS / Linux / Windows)

4. Find the Python that has the package installed

5. Write the Claude Desktop config file for you

Step 3: Restart Claude Desktop

Your 18 tools are ready. To get the most out of coaching, start by telling Claude your max HR and goals:

"Set my max HR to 190, resting HR 48, and my goal is sub-50 10K by June"

Claude will save this and use it to compute accurate training zones from then on.

Manual setup (if you prefer)

Add to your Claude Desktop config:

• macOS: ~/Library/Application Support/Claude/claude_desktop_config.json

• Windows: %APPDATA%\Claude\claude_desktop_config.json

• Linux: ~/.config/Claude/claude_desktop_config.json

{
  "mcpServers": {
    "strava": {
      "command": "strava-mcp",
      "env": {
        "STRAVA_CLIENT_ID": "your_client_id",
        "STRAVA_CLIENT_SECRET": "your_client_secret",
        "STRAVA_REFRESH_TOKEN": "your_refresh_token"
      }
    }
  }
}

Usage Examples

Once connected, just talk to Claude:

• "What's my current Z2 pace and how has it changed over the last 3 months?"

• "My left knee is sore — adjust my training plan for this week"

• "Am I on track for my sub-50 10K goal?"

• "How was my running this month compared to last month?"

• "Which shoes have the most miles on them?"

Memory is stored in ~/.strava-mcp/memory/ as plain JSON files — easy to inspect or back up.

Security

Credentials are loaded from environment variables only — never hardcoded. The server uses Strava's OAuth2 refresh token flow and automatically handles token renewal.

Requirements

• Python 3.10+

• A Strava account with API access

• Claude Desktop (or any MCP-compatible client)

License

MIT