WakaTime MCP Server

A Model Context Protocol (MCP) server that provides high-signal coding analytics from your WakaTime data.

• Direct mode: FastMCP serves Streamable HTTP at http://localhost:8000/mcp
• Proxy mode: mcp-proxy exposes the server over SSE/HTTP and Caddy adds token auth (recommended for self-hosting)

Tooling / API

Configuration

Configure via environment variables (or a .env file for the self-hosted scripts).

Development (Direct mode)

1. Install (uv required)

   git clone https://github.com/dpshade/wakatime-mcp.git
   cd wakatime-mcp
   
   uv sync --no-install-project
   

2. Run

   WAKATIME_API_KEY="your_wakatime_api_key_here" uv run -- python src/server.py
   

3. Connect

   • URL: http://localhost:8000/mcp
   • Auth: none

Deployment

Option 1: Self-hosted with auth (recommended)

This mode runs the MCP server with FastMCP’s default stdio transport and uses mcp-proxy to expose it over HTTP:

Internet -> (optional Tailscale Funnel) -> Caddy (auth) -> mcp-proxy -> FastMCP (stdio)
                                   :8770         :8767

1. Configure

   cp .env.example .env
   # Edit .env: set WAKATIME_API_KEY and a strong MCP_AUTH_KEY
   

2. Download Caddy (auth proxy)

   curl -L https://github.com/caddyserver/caddy/releases/latest/download/caddy_linux_amd64 -o deploy/caddy
   chmod +x deploy/caddy
   

3. Start

   ./deploy/start.sh
   

4. Endpoints

   • Auth proxy (recommended):
     • SSE: http://localhost:8770/sse
     • Streamable HTTP: http://localhost:8770/mcp
   • Internal (no auth; do not expose publicly):
     • SSE: http://localhost:8767/sse
     • Streamable HTTP: http://localhost:8767/mcp

mcp-proxy also exposes a health endpoint at http://localhost:8767/status (and via auth proxy at http://localhost:8770/status).

Systemd (persistent)

./deploy/install-systemd.sh
sudo systemctl enable --now mcp-wakatime mcp-wakatime-auth

Optional: Tailscale Funnel

If you use Tailscale, you can publish the auth proxy port:

tailscale funnel --bg --set-path=/wakatime localhost:8770
tailscale funnel --bg 443 on

Option 2: Docker

Runs mcp-proxy + the server in a container.

cd deploy
docker-compose up -d

• Endpoint (no auth): http://localhost:8767/sse
• If you want auth, run Caddy on the host (or add it to your own compose stack) and proxy to 8767.

Option 3: Render

This repo includes render.yaml for deploying the direct Python server.

• Set environment variable: WAKATIME_API_KEY
• Your service endpoint will be: https://<your-service>/mcp

Client setup

MCP Inspector

npx @modelcontextprotocol/inspector

Then connect using:

• Direct mode: http://localhost:8000/mcp (Streamable HTTP)
• Proxy mode: http://localhost:8770/sse (SSE)

Poke / other hosted clients (proxy mode)

Use the auth proxy SSE endpoint and send MCP_AUTH_KEY via one of:

• Authorization: Bearer <MCP_AUTH_KEY>
• X-API-Key: <MCP_AUTH_KEY>
• Api-Key: <MCP_AUTH_KEY>

Security notes

• Generate a strong auth key:

  openssl rand -hex 32
  

• Never expose the unauthenticated mcp-proxy port (8767) to the public internet.

License

MIT