Someone probed McDonald's MCP so you don't have to

What it does

M-CODE WRAPPED is a Python toolbox built on top of McDonald’s China MCP server. It has four parts: a capability explorer that calls every read-only tool and records what actually comes back, a personal annual consumption report (orders, spend, points, late-night order share, top items), a public-data panorama covering 158 menu items’ nutrition plus marketing calendars and coupon pools, and a Monte Carlo expected-value calculator for the points lottery. There’s also optional voice narration in Cantonese, Hokkien-accented Mandarin, or plain Mandarin.

The interesting bit

The explorer exists because the official docs are thin, and the reality is messy: responses are Markdown, not JSON, with the real payload hidden after an ## Original Response header; nutrition data comes back as comma-separated text tables; and empty results are Chinese prose like 暂无可用优惠券 instead of empty arrays. The server actually exposes 35 tools, not the 33 the docs claim. That’s the kind of field report anyone integrating this MCP server will want before writing a single line of glue code.

Key highlights

• Explorer auto-chains parameters: storeCode from order-list feeds query-meals, orderId feeds query-order, and so on — 27 of 35 tools successfully probed
• Hard safety line: all write operations (draw-lottery, create-order, cancel-order, etc.) are never called, only documented
• Lottery EV uses the real prize pool and the real 24-point draw cost, with coupon face values matched to base prices (McFlurry at ¥9, combo at ¥40)
• Zero third-party dependencies for the explorer script
• Includes a SKILL.md so an agent can drive the whole pipeline

Caveats

• McDonald’s doesn’t publish lottery win probabilities — all EV results assume equal probability, which the README discloses but which still makes the numbers speculative
• 7 of 35 tools went unprobed (the write ops), and one failed for missing parameters, so coverage isn’t total
• It’s a hackathon entry (1024 programmer festival contest), not an official product — treat it as a well-documented experiment

Verdict

If you’re integrating the McDonald’s MCP server, the explorer report alone is worth the click — it’s reverse-engineered documentation the official docs should have been. If you just want cheaper burgers, this isn’t your app, and it deliberately isn’t trying to be.