Tesla's in-car browser — like most vehicle infotainment browsers — refuses to play a <video> element while the car is in drive. That's a deliberate safety restriction, enforced by the browser itself, and no app can legitimately talk its way around it. So MirrorCast doesn't try to. It does something different: it never sends video at all.
Instead of capturing your phone's screen as a video stream, MirrorCast's Android capture pipeline grabs individual snapshots of the display — many times per second — using the system's screen-capture API. Each snapshot is a plain still image. It gets downscaled (to 900, 1280, or 1920 pixels on the longest side, depending on the quality you pick) and compressed as a JPEG at around 50–55% quality. That single compressed image becomes "the current frame," held in memory, overwritten by the next one a fraction of a second later.
At no point does this pipeline touch a video codec, a video container format, or anything a browser would recognize as "video." It's a sequence of ordinary photographs, produced roughly twelve times a second.
When the car's browser opens the link MirrorCast gives you, what loads is a completely ordinary <img> tag. Not a <video> tag — an image tag, the same kind used for a logo or a photo on any webpage. Behind the scenes, that image's source gets swapped out for a fresh one roughly twelve times a second. As far as the browser's safety logic is concerned, there is no video anywhere on the page. There's a picture. It simply doesn't know, or care, that the picture keeps changing.
This is the entire trick: MirrorCast never crosses the line the browser is watching for. It's not circumventing the video lock — technically, there's no video for the lock to catch.
Human vision blends a fast enough sequence of still images into the perception of smooth movement — that's the same basic principle behind traditional film (24 frames per second) and, further back, the flipbook. MirrorCast doesn't need to hit true video frame rates to feel live; twelve frames a second is enough for menus, maps, videos, and most on-screen content to read as fluid motion rather than a slideshow.
Being honest about the tradeoffs matters here. Real video codecs (H.264, HEVC, etc.) compress footage by only encoding what changed between frames — that's why a two-hour movie fits in a few gigabytes. MirrorCast's approach sends each frame as a fully independent JPEG with none of that inter-frame compression, which is fundamentally less bandwidth-efficient. That's exactly why the frames are downscaled and JPEG-compressed fairly aggressively — to keep the stream smooth rather than saturating your connection.
Audio travels alongside the images and stays in sync with them on the device you're watching from, so picture and sound arrive together rather than audio simply playing out of your phone's own speaker. This covers any app on your phone — YouTube, Spotify, maps with voice turn-by-turn, even a browser tab — not just a specific allowlist.
Because that would run straight back into the same wall. Any approach that delivers genuine video — whether that's a raw H.264 stream, WebRTC, or wrapping frames back into a video container — hands the browser something it can correctly identify as video, and Tesla's safety restriction would apply to it exactly the same way it applies to Netflix or YouTube's own video player. Staying on the "sequence of still images" side of that line isn't an implementation detail; it's the whole reason this works at all.