Your food diary, written by your assistant

ChatGPT, Claude and coding agents get the same Gutlog you use on the web: log a meal from a photo, fix it, read the day back and, as they ship, check how clean a product is and look for triggers. Over MCP, the CLI or REST.

Set it up Read agents.md

claude · gutlog mcp

[photo] Log my breakfast.

log_meal({ "items": [{ "name": "scrambled eggs", "grams": 120, … }, { "name": "sourdough toast", … }] })

352 kcal · 19.8 g protein · egg confirmed · butter 55%, milk 35% to confirm

Logged: scrambled eggs on toast, 352 kcal, 20 g protein. Were the eggs cooked in butter? Any milk in them?

How am I doing today?

get_day()

1 meal · 352 kcal · 19.8 g protein · 0 to review

One meal so far: 352 kcal and 20 g protein. Nothing waiting for review.

What makes it agent-friendly

One tool per op

Every web app action is an MCP tool with a description written for a model, plus read-only and destructive hints, so clients can auto-approve reads and confirm the rest.

The host does the vision

The assistant already sees the photo, so it sends log_meal a structured draft. No upload, no second model, and logging is free and unlimited.

Confidence the model can act on

Every ingredient comes back in a tier with its confidence, plus at most three questions to ask. The model never has to guess what to confirm.

Rules for health data

agents.md tells assistants never to present a gut pattern as a diagnosis, to run a program's safety screen first and to keep what you eat inside the conversation.

A person approves money

Assistants read prices and hand you a checkout link; changing the plan on a saved card needs an explicit confirmation and its own scope.

Readable by machines

setup.md, llms.txt, llms-full.txt, agents.md, OpenAPI 3.1 and an MCP server card describe the whole API, generated from the same registry that serves it.

A plugin with skills

The plugin for Claude Code, Codex and Cursor (/plugin marketplace add MeterApp/gutlog-plugins) bundles the MCP server with skills for logging food, reviewing the diary, gut checks and clean checks.

Errors an agent can act on

RFC 9457 problems with a code, a request id and, for plan limits, a checkout link. Issues can be reported with the failing request attached.

Connect in one command

Claude Code, signing in with your browser. Or paste one sentence into ChatGPT or Claude: the setup page has it.

claude mcp add --transport http gutlog https://getgutlog.com/api/mcp

ChatGPT, Claude, Codex, Cursor, VS Code, Gemini CLI and API keys are on the setup page. Prefer plain HTTP? Every tool is also POST /api/v1/{op}.

Your next meal is the first data point

Paste one sentence into ChatGPT or Claude and your assistant sets it up. Free to start, no card, and logging from your assistant is unlimited.

Set it up Read the quickstart