An MCU (Multipoint Control Unit) is a media server architecture in which every participant uploads their stream to the server, the server fully decodes all incoming streams, composites them into a single mixed video output — typically a grid or speaker-focus layout — and then re-encodes and delivers that composite to each participant as a single stream. This contrasts sharply with an SFU, which merely forwards individual streams without any decoding or re-encoding. The benefit of an MCU is that each participant sends one stream and receives one stream regardless of how many people are in the call, which minimizes bandwidth at the client side — historically important when browser and connection capabilities were weaker. The major cost is server-side: decoding, compositing, and re-encoding N simultaneous streams is CPU-intensive and scales poorly as participant count rises, making MCU infrastructure expensive to run at live-class scale. The composite layout is also fixed at the server, so the MCU cannot adapt to individual viewer preferences — one viewer cannot pin a specific speaker without the server rendering a separate composite for them. Modern live-learning platforms predominantly use SFU architectures because client devices are now capable of handling multiple decoded streams, and simulcast plus SVC provide server-side quality adaptation without transcoding. MCU is still encountered in legacy conferencing systems and in scenarios where extremely limited client bandwidth makes a single received stream mandatory, such as very low-end mobile devices on 2G networks.
