When would you choose an MCU over an MPU (or vice versa)?

Choose an MCU when the job is real-time control, low power, low cost, and instant-on with a small, fixed software footprint, motor control, sensor acquisition, a battery device that sleeps in microamps, anything with hard deadlines where deterministic timing matters. Choose an MPU when you need an operating system and its ecosystem: a rich GUI, a networking stack with TLS, a filesystem, multiple concurrent applications, or heavy compute like video or ML. The tradeoffs are concrete, the MPU buys you capability at the cost of higher power, a larger bill of materials (external DRAM, storage, PMIC), seconds-long boot from loading an OS, and harder-to-bound worst-case timing due to caches, MMU, and OS scheduling. A common architecture splits the difference: an MPU runs the application while a companion MCU (or an MCU core on the same SoC) handles the hard-real-time, low-power tasks.