Supercut + Mixture
Supercut is where the recording happens. Mixture is where it becomes a video someone would watch twice.
They pair well because they solve opposite halves of the same job: Supercut captures the session — the screen, the camera, and, crucially, the data about what happened — and Mixture turns that into a directed piece with framing, motion and pacing you control. Neither has to guess what the other meant, because an agent connected to both moves the material across for you.
The setup
Section titled “The setup”Connect both MCP servers to the same agent — Mixture’s takes one command (full setup), and Supercut’s is on their side. From then on your agent can read your recordings and build in your canvas in the same breath:
“Grab my latest Supercut recording and cut it into a 45-second product demo — our usual treatment.”
First-class ingestion is on the way. Today an agent wired to both is already the bridge, and it’s a one-line ask.
What Mixture adds to a capture
Section titled “What Mixture adds to a capture”A raw screen recording is not a demo. These are the things the finish is actually made of — all ordinary layers and animations you can retime afterwards, not a filter applied over the top.
A real cursor, not the baked-in one
Section titled “A real cursor, not the baked-in one”The pointer in a screen capture is pixelated, jittery, and stuck at whatever size the OS drew it. If Supercut captured the mouse data, hand the agent that file — a CSV or JSON of positions over time — and Mixture drives a clean overlay cursor along the real path instead.
The difference is not cosmetic. A data-driven cursor can be scaled for a vertical crop, eased so it arrives instead of snapping, given a click animation, hidden while it’s irrelevant, and re-timed when you retime the clip. It’s a layer bound to data, which means everything in data-driven video applies to it.
Frame-accurate cuts, without the wait
Section titled “Frame-accurate cuts, without the wait”A screen capture is exactly the kind of file that’s painful to cut: long, and storing a complete picture only every few seconds. So the first time Mixture sees one it builds an editing-optimised copy locally — you’ll see an Optimising source files percentage while it works, once per recording, in your browser, with nothing uploaded. After that, cuts, speed changes and punch-ins land on any frame in about a second. How that works.
Punch-ins that follow the action
Section titled “Punch-ins that follow the action”Zooms are directed, not global: each one has a focus point, a level, and its own ramp in and out. The agent places them on the moments that matter — the click, the result, the number changing — and you drag them on the timeline afterwards if the timing isn’t quite right.
Chained punch-ins glide between focus points without zooming out in between, which is what makes a sequence feel like camera work rather than a slideshow.
With the recording’s mouse data attached, a punch-in can follow the cursor: the view pans as the pointer nears its edge, late and soft, and holds still while you read. It’s a toggle on the zoom, on by default for recordings that have the data, and Auto zooms on the clip menu proposes punch-ins from the clicks for you or the agent to prune.
Motion blur that matches the move
Section titled “Motion blur that matches the move”Fast motion without blur reads as stutter. Mixture’s motion blur has modes — radial from centre, directional, or following the measured motion of the layer — so a whip-pan looks like a whip-pan and a punch-in looks like a lens moving, not a frame skipping. More on the modes.
The frame around it
Section titled “The frame around it”The conventions that separate a launch video from a screen grab: a squircle screen frame with the right corner radius for the platform, a dark backdrop with a scrim, layered shadows that read as depth, and your camera clip as a circular bubble in the corner when you recorded one.
A camera you direct
Section titled “A camera you direct”The composition itself has a camera — pan, zoom and rotation you can animate across a whole scene, on top of the per-clip punch-ins. That’s how you get a slow drift across a dense UI, or a hold that eases as the voice-over lands.
Everything else, once it’s in
Section titled “Everything else, once it’s in”Captions cut from the audio with word-level timing, a voice-over placed against the edit, music ducked under the narration, a 9:16 cut reframed properly rather than centre-cropped — the rest of the toolkit applies to a Supercut recording exactly as it does to anything else.
Ask for it in one line
Section titled “Ask for it in one line”“Turn the Supercut recording into a demo: screen frame on a dark backdrop, punch in on each click, drive the cursor from the mouse data, and add captions.”
“Same again for this week’s build — use our Product Demo recipe.”
That second one is the point of doing it twice. Once the treatment is right, keep it as a recipe and every future recording gets the same finish from a one-line ask — the footage slot fills with the new capture and everything else holds.
Then take it apart
Section titled “Then take it apart”Nothing an agent builds is special: the zooms, the cursor, the frame and the blur are all ordinary layers, animations and effects. Retime a punch-in on the timeline, restyle the backdrop, swap the music, or delete the cursor entirely and drag your own. The agent gets you to a finished video; the canvas is there for the last ten percent that makes it yours.