An MCP skill that price-checks your McDonald's order before you order it

What it does

This is an AI-client “skill” (tested mainly with WorkBuddy, but any remote-MCP client works) that takes a budget and a craving — say, ¥30, one burger, one drink — queries your chosen store’s menu, coupons, and combo details, then runs up to three candidate meals through McDonald’s official calculate-price endpoint. You get the real payable total, what’s actually in the bag, remaining budget, and coupon conditions. It stops at recommendations; it never places an order.

The interesting bit

The discipline around pricing is the whole point. Every menu change triggers a fresh official quote — swap the cola for a sugar-free one and the entire order is re-priced, never patched with menu arithmetic. A local Python script (stdlib only, no dependencies) then sanity-checks each quote down to the integer fen: budget fit, item counts, exclusions, quote freshness, duplicates. It’s glue code, but unusually honest glue.

Key highlights

• Real quotes from the official McDonald’s China MCP (https://mcp.mcd.cn), including a documented case where a drink swap still totaled ¥22.90 — verified, not inferred
• Token-free offline demo pages let you play with budget sliders and historical (sanitized) quotes before installing anything
• Failure handling is thoughtful: if a later request fails, earlier successful quotes are kept with the failure reason; reports can be regenerated offline from saved JSON
• Private data stays private — raw requests/responses land in a git-ignored private/ directory, and public examples scrub store names and timestamps
• 51 offline regression tests for the Python logic, 14 DOM scenarios for the report pages

Caveats

• Single store, single person, dine-in pickup only; comparisons cover only the current round’s candidates, not the whole menu’s lowest price
• The coupon-enabled branch has no real-account test case yet, and WorkBuddy skill auto-discovery is unverified
• Nutrition, ingredients, and allergens are explicitly out of scope

Verdict

A charming, unusually rigorous example of what an MCP skill can be — worth a look if you build agent tooling or just want your ¥30 to go further. If you were hoping for nutrition guidance or full-menu optimization, it stops short of both, by design.