Healthy MCP template

A private health helper template to use with your favorite AI assistant that supports MCP servers. Say what you ate, drank or did; the assistant calls this server; the server stores it and does all the nutrition math. It is built to be your own copy: you run it on your own Cloudflare account and your data stays there.

This is a template. Click Use this template at the top of the GitHub page (or use it directly) to get your own copy of the code in your own GitHub account, then follow the steps below. Nothing in it belongs to anyone's account: no ids, names or URLs are stored in the repo, the list of people who may sign in starts empty, and every key is yours and lives in your GitHub secrets.

Set up your MCP

You need a GitHub account and a Cloudflare account. The free plans are enough to start (check each provider's current limits).

Throughout, OWNER/REPO means your copy of this repo, for example redSlug/healthy-mcp-template.

1. Use this template

Click Use this template > Create a new repository (you can make it private). Or clone or fork this repo.

2. Cloudflare: account id, subdomain and API token

1. Sign up and create an account (best practice: a separate account for each MCP server you run), then open Workers & Pages. Pick a workers.dev subdomain if asked, and note it and your Account ID (on the right).

2. Under My Profile, API Tokens, create a custom token for your account only, with Edit permission for Workers Scripts, D1 and Workers KV Storage, and Read for Account Settings.

3. GitHub OAuth App (this is how people sign in)

GitHub cannot create these automatically, so do it once by hand: github.com/settings/applications/new.

Leave the checkboxes Allow wildcard matching, Enable Device Flow and Expire user access tokens unchecked: the server uses one exact callback address and the normal browser sign-in, and it uses the GitHub token once to read who signed in and never stores it.

Click Register application, then Generate a new client secret. Copy the Client ID and the client secret.

4. USDA food database key

Get a free key at fdc.nal.usda.gov/api-key.html. It is emailed to you.

5. Put the keys into your GitHub repo

Fill in the values at the top of scripts/sample_script.sh, including the GitHub usernames of the people allowed to sign in, and run it with bash scripts/sample_script.sh. It also prints your server address. Then discard your edits so the keys are never committed.

Who may sign in: edit config/allowed-users.json to a list like [{ "github_id": 12345678, "login": "your-login" }]. Find an id at https://api.github.com/users/ plus the username (for example api.github.com/users/redSlug), under id. Commit it.

Secrets, at https://github.com/OWNER/REPO/settings/secrets/actions (New repository secret):

Variables, at https://github.com/OWNER/REPO/settings/variables/actions (New repository variable, not secrets):

6. Deploy

Go to https://github.com/OWNER/REPO/actions/workflows/deploy.yml, click Run workflow, then Run workflow again. When it turns green (a minute or two), your connector address is WORKER_URL/mcp.

1. runs the tests;

2. checks that every key and setting above is present;

3. creates your D1 database and KV storage in your Cloudflare account (only what is missing);

4. sets up the database tables;

5. deploys the server together with its secrets;

6. waits for the server to come online (a brand-new one can take a few seconds) and checks that it refuses requests without a token.

Related MCP server: mcp-nutrition-db

Use your MCP

Add WORKER_URL/mcp (it must end in /mcp) as a remote MCP server in your assistant, and sign in with GitHub when asked. Then try: "Log 150 grams of cottage cheese for lunch", "I drank a glass of water", or "I ran for 30 minutes".

1. Open claude.ai/settings/connectors and choose Add custom connector.

2. Enter WORKER_URL/mcp. Choose Sign in now, and let the OAuth client register automatically. Leave headers empty.

3. Approve the screen and sign in with GitHub.

4. In a new chat, turn the connector on.

More: day to day, updating your copy, troubleshooting, how it works and local development.

License

MIT