AI-powered fitness and nutrition tracking. Works locally via stdio (single user) or deployed as an HTTP service with multi-user auth.

EchoFit is organized into feature modules that can be enabled or disabled independently:

Modules share a common config and user identity layer but do not depend on each other.

Modules can be enabled at the server level or per-user. A deployment can expose all modules or a subset — for example, a "calorie counter" product surface that only enables diet tracking in calorie-only mode, while a "full fitness tracker" enables everything.

When a module is disabled for a user, its MCP tools are excluded from tools/list responses — the AI agent never sees them and never mentions them. This is controlled via:

• Server-level: Configuration in the deployment determines which modules are active for that instance
• Per-user: JWT claims can scope a user to specific modules, allowing different tiers or product surfaces from the same server

This means you can publish multiple products (e.g., "EchoFit Calorie Counter", "EchoFit Pro") to different connector stores, each pointing at the same server but with different module configurations via the client credentials or URL path.

Module toggling is a runtime concern, not a packaging or deployment concern. Every deployment includes all modules — gapp deploy produces the same artifact regardless of which product surface you're serving. What varies is the server configuration and JWT claims that control tools/list visibility.

pip install -e sdk/ -e mcp/

Claude Code:

claude mcp add echofit -- echofit-mcp stdio --user local

Gemini CLI:

gemini mcp add echofit -- echofit-mcp stdio --user local

Start a conversation with your AI assistant and ask it to log food, show your food log, manage your catalog, etc.

For remote access from Claude.ai, mobile, or multiple devices, deploy as an HTTP service. The server uses mcp-app for server bootstrapping and user-identity middleware.

gapp deploys to Google Cloud Run with infrastructure, secrets, and GCS FUSE data volumes:

gapp init
gapp setup <project-id>
gapp deploy

See gapp documentation for details.

Any platform that runs Python ASGI apps works — Docker, Fly.io, Railway, etc. Set the environment variables above and run with uvicorn or any ASGI server.

The echofit-mcp package ships an echofit-admin CLI for registering users, issuing tokens, and checking service health. It talks to the /admin REST endpoints on the running server.

# Point the admin CLI at your deployment
echofit-admin connect https://YOUR-SERVICE-URL --signing-key "$SIGNING_KEY"

# Register a user and get their token
echofit-admin users add alice@example.com
echofit-admin tokens create alice@example.com

# List / revoke
echofit-admin users list
echofit-admin users revoke alice@example.com

# Health check
echofit-admin health

The token printed by tokens create is what the end user puts in their MCP client config (see below).

For local development against the filesystem store, use echofit-admin connect local instead of a URL.

Claude Code / Gemini CLI (Authorization header):

{
  "mcpServers": {
    "echofit": {
      "url": "https://YOUR-SERVICE-URL/",
      "headers": {
        "Authorization": "Bearer YOUR_TOKEN"
      }
    }
  }
}

Claude.ai (query param):

https://YOUR-SERVICE-URL/?token=YOUR_TOKEN

sdk/echofit/app.yaml configures when the calendar day rolls over:

hours_offset: 4
timezone: "America/Chicago"

hours_offset: 4 means eating at 2 AM counts as the prior day (the new day starts at 4 AM). Adjustable for any schedule.

Local data follows XDG conventions:

• Data: ~/.local/share/echofit/
• Config: ~/.config/echofit/

Override with ECHOFIT_DATA and ECHOFIT_CONFIG env vars.

See CONTRIBUTING.md for architecture, testing conventions, and how to add features.

python -m venv .venv && source .venv/bin/activate
pip install -e sdk/ -e 'mcp/[test]' -e cli/
python -m pytest

The [test] extra pulls in pytest, pytest-asyncio, and httpx — required to run the free mission-critical tests that the mcp-app framework provides against this solution. See CONTRIBUTING.md for what those tests verify.

• mcp-app — MCP server framework with user-identity middleware
• mcp — MCP protocol SDK
• PyYAML — configuration