An engineer reverse-engineered McDonald's app to map every broken ice cream machine

McDonald's soft-serve machines are infamously prone to breaking down, and in 2020 customers had no way to know a machine was down short of driving to the store. That October, Germany-based software engineer Rashiq Zahid launched McBroken, a map that colors every US McDonald's green or red depending on whether its ice cream machine works. His method: reverse-engineer the chain's internal ordering API and attempt to add a McSundae to a cart at every location. If the app accepts the item, the machine is up; if it refuses, it is down.

The probe needed tuning. Polling every minute made the app peg him as a bot and cut off access; adding a McSundae every 30 minutes keeps the map current and passes its human-seeking standards. Zahid started with McDonald's locations in Germany, biking between Berlin stores to check McBroken's data against reality, and only then expanded to the US. The picture it painted was blunt: just under 10 percent of US machines broken nationwide at the time of writing, nearly 24 percent in New York City, about 20 percent in Seattle and roughly 14 percent in the Washington, DC area.

An earlier attempt, Raina McLeod's 2017 app built after an unfulfilled McFlurry craving, relied on crowdsourced reports and was only as good as its users; querying the ordering source directly made McBroken's data more accurate and more current. McDonald's did not fight the project — communications and government relations executive David Tovar publicly thanked Zahid, calling him a true fan and conceding the chain had 'opportunities' on sweet treats — and Business Insider reported franchisees were hunting fixes of their own.

Don't crowdsource outage reports — probe the ordering API. If the app lets you add an ice cream to the cart, the machine works; if it refuses, it's broken.

The ordering app is a de facto machine-status API: its cart logic already knows which locations cannot serve ice cream.

A synthetic cart costs nothing and is always fresh, unlike crowdsourced reports that decay with user attention.

The 30-minute cadence was the pivot — fast enough to stay current, slow enough to look human to the app.

The map shames without accusing: it simply states what the ordering app already implies about each store.

When a company won't publish status data, its own transaction interface still leaks it. One well-chosen probe that fails exactly when the thing is broken beats an army of human reporters.

McDonald's never moved against the bot: David Tovar applauded the effort publicly ('Only a true McDonald's fan would go to these lengths') and acknowledged the company had opportunities to satisfy more customers on sweet treats. Business Insider reported the same month that franchisees were taking the matter into their own hands and seeking solutions to keep machines working more often.