Describe the shot, get the scene: AI composition and the Mockpose MCP server

Most "AI design" tools hand you a picture. You type a prompt, you get a flat image — and the moment you want the phone tilted two degrees, the background a touch darker, or the headline in another language, you're rolling the dice again. It looks like a scene, but there's nothing behind the pixels to edit.

Mockpose takes a different path. Our AI doesn't paint a picture of a device mockup — it composes a real 3D scene, the same kind you'd build by hand in the editor. Every device, camera move, light, and animation it chooses is a live object you can select, tweak, or replace. Nothing is baked in.

From one sentence to a full scene

Describe the shot in plain words:

An iPhone spiralling in, dark studio, slow push-in.

That's enough. Mockpose composes the complete scene — devices, camera move, lighting, background, animations — and drops you into the editor with everything live. Push-in too fast? Drag a keyframe. Studio too moody? Raise a light. The AI's choices are starting points, not verdicts.

It works on existing scenes too. Give an instruction — "make the background warmer", "have the phone enter from the left" — and the AI edits the scene you're already working on instead of starting over.

Why this works: the scene is a document

The trick is structural, and it's the best kind of boring: a Mockpose scene is one structured JSON document. Devices, camera path, lights, background, animation tracks, localized text — everything lives in a single well-defined format that the editor reads and writes on every click and drag.

So instead of generating pixels, our AI writes that document. There's no translation layer and no lossy hand-off: the scene the AI composes is exactly the scene you open. That's the difference between an AI that draws your idea and an AI that builds it.

Claude at the controls: the MCP server

The same document model powers our remote MCP server. Connect Claude — or any MCP-capable agent — and it can drive the whole studio through tools:

• Add devices and apply templates
• Animate the camera and the devices
• Set backgrounds and lighting
• Manage localized text across languages
• Check the layout and render the output

Because a human dragging a device and an agent calling a tool write to the same scene document, there's no separate "agent mode". Whatever an agent builds, you can open and polish by hand — and whatever you build by hand, an agent can pick up and extend.

Three ways to put it to work

1. Release automation. An agent that turns every release into a fresh preview video — new screens in, camera move applied, output rendered, no manual steps.
2. Localized variants in batch. Produce one scene in every store language using the built-in localized text. If that's your bottleneck, start with our guide to localized App Store screenshots.
3. A bigger pipeline. Wire Mockpose in next to your other MCP tools, so device visuals become one step in your content workflow instead of a manual detour.

The practical details

AI scene generation and the MCP server are premium features, and your monthly AI quota is visible right in the app — no surprises. Underneath them sits the same studio everyone gets: 100+ editable templates, cinematic camera paths, a keyframe timeline, and export up to 4K/60 fps. The free plan covers the full editor with 720p watermarked export, so you can get comfortable with scenes before an AI ever writes one for you. New here? Start with the beta announcement.

Describe the shot. Get the scene. Keep every choice editable. Open app.mockpose.com and try it — and if you have an agent handy, point it at our MCP server and let it do the dragging.