Topic → script → voice → on-brand animated explainer → SEO pack → scheduled post. Drive it yourself in a visual editor, or connect your AI agent and let it run the channel — both edit the same project, rendered by the same engine.
The video on the right was rendered by our own engine — no stock footage, no generative clips. A character, cards and motion that stay identical across every video you make.
Most tools give you either an editor or an API. We built one project model with two drivers — so anything your agent makes, you can open and fix by hand, and anything you make, your agent can continue.
Scenes on the left, a live animated preview in the middle, every element's text, position, timing and entrance motion on the right. Drag the timeline, change a headline, re-time a scene — the preview is the real engine evaluating the real project, in your browser.
Claude, ChatGPT or any MCP client connects to the same project surface: 23 tools to create projects, add scenes and elements, write narration, fit timing to the voice, preview frames and render finished MP4s. Named features are defaults, not ceilings — primitives compose freely.
From an idea to a posted video, without a timeline editor, a render farm, or a posting spreadsheet.
Start from a proven format — the cartoon-teacher + cards explainer that fills faceless feeds — or compose scenes from primitives: text, cards, bars, pills, characters, diagrams.
Narration is generated from your script, and scene timing is fitted to the voice automatically — elements enter on the words they belong to.
Every scene, element, entrance and hold motion is editable in the browser editor, and equally editable by your agent through MCP. Same file, same result.
The deterministic engine renders the finished MP4 — the character, spelling and layout hold steady across every shot, something generative video still can't do. SEO pack, scheduling and publishing are built in.
Every video below came out of the same engine and project model you can edit in the browser — no hand assembly.
The Attention Trap — the full 52s template: 24 shots, one rigged character, hooks, steps, open loops, CTA. Open it in the editor →
Hooks demo — motion system proof — the agent-built primitives project, reworked by the engine's v3 motion system: nothing sits frozen. The card floats while it holds, the bar shimmers, the pill pulses, the icon spins slowly — and the card and pill don't enter on a timer, they enter on the narration words they belong to ("curiosity", "question"). Leo only talks while his scene's narration is actually speaking. Open it in the editor →
Not a "generate" button — the whole studio, as composable tools. An agent can build a video it has never seen a template for, check its own frames, fix them, and render.
| Tool | What it does |
|---|---|
| get_template_info | Template schema, full example payload, shot list, poses, audio note |
| create_project / get_project / list_projects | Project lifecycle on the shared project model |
| add_scene / update_scene / remove_scene | Scene order, duration, background |
| add_element / update_element / remove_element / list_elements | Text, cards, pills, bars, characters, shapes, icons, images, progress dots — position, size, timing, style, enter / hold / exit animations |
| stagger_scene / resolve_sync / validate_project | Stagger a scene's entrances in one call; resolve narration-synced entrances to real times; surface warnings (stacked entrances, static stretches, broken syncs) before rendering |
| set_narration / fit_timing_to_narration | Script → generated voice; scene durations fitted to the narration |
| template_to_project | Turn a template into a fully editable project |
| render_preview_still / render_project_still | One frame at any second (~1s) — the agent checks its own work before rendering |
| make_video / render_project / get_render_status / list_renders | Finished MP4 renders, job status, render library |
Priced on the outcome — a running channel — not on credits. Our render costs cents, so allowances can be generous.
Launch pricing from the product plan; checkout opens with hosted accounts (next phase). The editor above is live and free to try today.
The MCP server runs locally alongside the engine today (hosted MCP with OAuth is the next phase). Point any MCP client at it and your agent gets all 23 tools.
{
"mcpServers": {
"attention-trap-explainer": {
"command": "node",
"args": ["<path-to>/explainer-engine/mcp/dist/index.js"]
}
}
}
Or with Claude Code:
claude mcp add attention-trap-explainer -- node <path-to>/explainer-engine/mcp/dist/index.js
Ask it to make the same video — it reads the template, previews stills, renders the MP4. Or give it a topic: it creates a project from primitives, writes the narration, fits the timing, checks its own frames, and renders a finished video it designed itself.
On Pro, the same connection runs campaigns: plan a month of videos, produce them on schedule, and report back — you just approve.