Summary

Useful only if you already have access to the system behind it.

The relative-period argument is the sensible design choice — "last week" resolves server-side, so you never hand the model a date it computed wrongly. But the API base URL and token both come from an administrator, which makes this an internal tool published openly rather than something you can try. Read it as a compact example of wrapping a private endpoint if that is the shape you need.

What it is

A small Python server that fronts an internal canteen-statistics API. One tool answers how many people ate breakfast and lunch over a window, either as explicit dates or as a named relative period, and returns the two counts plus a total as readable text.

What you get

`get_canteen_data` returns breakfast, lunch and total headcounts
`start_date` and `end_date` in `YYYYMMDD` form for an explicit window
Or a `period` shortcut: `today`, `yesterday`, `day_before_yesterday`, `this_week`, `last_week`, `this_month`, `last_month`
Results come back as formatted text rather than raw JSON

Requirements

Two environment variables you cannot self-serve: `CANTEEN_API_TOKEN` and `CANTEEN_API_BASE`, both obtained from whoever administers the canteen system. Install from PyPI as `mcp-server-canteen`, version 0.2.8, and run it through `uvx` from your client config. Without a valid token and base URL the server starts but every call fails.

Setup effort

One command plus a key — pip install mcp-server-canteen, then supply credentials

Canteen Data