KynimoGitHub

Motion graphics for your AI agent.

Kynimo is an MCP server that gives Claude, Cursor or any assistant a real motion engine. It builds the video from layers and keyframes, looks at the frames it made, and hands you a project you can still change.

Follow on GitHubConnect your assistant
launch-teaser1920 × 1080 · 30 fps · 4 layers00:03:10 f100
markshape
titletext
ruleshape
captiontext

A scene document, drawn live in your browser. Drag the timeline.

Not a clip. A project.

Video generators hand you finished pixels. Want a different word? Generate again and hope.

Kynimo builds a scene: every layer, keyframe, colour and timing has a name. Your assistant changes one of them, and every other frame stays exactly as it was.

  • background: solid
  • mark: shape · 6 paths
  • title: text · per-word animator
  • rule: shape · trim path
  • caption: text

What your assistant does with it.

It works like a motion designer with a checklist: build, look, check, fix. Every step is a tool call you can read.

  1. f000

    Brief

    You say what the video is for, in plain words. A launch teaser, a feature promo, a 10 second social clip.

  2. f030

    Build

    It lays out shapes, text, images and your footage, then animates them with spring and bezier keyframes.

    • add_layer
    • set_keyframes
    • apply_ops
  3. f060

    Look

    Rendered frames come back into the chat as images. Your assistant sees what it made, not what it meant.

    • render_frames
    • render_contact_sheet
  4. f090

    CheckNext

    The frames are graded against the style's rules. Each finding names the frame, the layer and the fix.

    • check_quality
  5. f120

    Fix

    Small typed edits to the scene. One layer changes, every other frame stays exactly as it was.

    • apply_ops
    • undo
  6. f150

    ExportNext

    MP4, WebM or GIF, and a link your team can scrub and comment on. The project stays editable.

    • export

One line to connect.

Kynimo runs in the cloud. Add the server, sign in once in your browser, and ask for a video. No install, no API keys to paste.

Terminal
claude mcp add --transport http kynimo https://mcp.kynimo.example/mcp

The hosted endpoint opens with the private beta. Until then the address is a placeholder.

Four styles. Four different films.

A style pack is a motion design system: palette, springs, staggers and timing. Each one gets a ten-second study here, drawn to show how differently the packs move.

Keynotespring 1.0 / 0.7 s · stagger 100 ms

Calm and confident. One idea at a time, long holds, springs that never bounce.

Playful SaaSspring 0.5 / 0.45 s · stagger 65 ms

Friendly and quick. Bright accents, tight staggers, springs that bounce once.

Editorialfade 800 ms · stagger 140 ms

Measured, like a magazine spread in motion. Light type, slow fades, a strict grid.

Brutalistlinear 200 ms · stagger 67 ms

Loud and blunt. Nothing eases in. Things appear and stop.

Built in the open.

No launch countdown. Here is what runs today and what comes next, in order.

  1. Scene format

    A versioned JSON document. Every change is a typed operation.

    Built.

  2. Engine

    Bezier and spring keyframes, parenting, precomps. Integer frames.

    Built.

  3. Renderer

    Shapes, text, images, video. Masks, mattes and blend modes.

    Built.

  4. Shape operators

    Trim paths, repeater, round corners, path morphing.

    Built.

  5. Text animators

    Per character, word and line, with range selectors.

    Built.

  6. MCP server

    Streamable HTTP with OAuth. Frames come back as images.

    Built.

  7. Style packs

    Four drafts: Keynote, Playful SaaS, Editorial, Brutalist.

    Built.

  8. Hosted endpoint

    One URL for every client.

    Planned.

  9. Intent tools

    Animate, stagger, retime. Your assistant says what, not how.

    Planned.

  10. Scripts

    Your assistant writes TypeScript. The scene gets ordinary layers.

    Planned.

  11. Quality report

    Every finding names the frame and the fix.

    Planned.

  12. Components and shaders

    TypeScript and GLSL, with typed props.

    Planned.

  13. Export

    MP4, WebM, GIF and a link your team can scrub.

    Planned.

Questions

What is Kynimo?

A cloud-hosted motion graphics engine built for AI assistants. Your assistant connects over MCP and builds a video out of layers, keyframes and components. You get a project you can keep editing, not a flat render.

Which assistants does it work with?

Any client that speaks the Model Context Protocol over HTTP. Kynimo is one remote server with OAuth sign-in, so there is nothing to install.

How is this different from AI video generators?

Generators return pixels. To change a word you generate again and hope. Kynimo returns a scene document, so a word, a colour or a timing is one small edit and every other frame stays the same.

What is a style pack?

A motion design system written as tokens and rules: type, colour, springs, durations, safe areas, reading time. Four draft packs exist today. Kynimo is being built to grade every render against the pack and tell the assistant exactly what to fix.

Can I use it today?

Not yet. The scene format, engine, renderer and MCP server are built and tested. The hosted endpoint, quality report and video export come next. The repository is public, so you can follow each step.

What will it make?

Short motion pieces, 5 to 60 seconds: product launch videos, UI promos, social clips and explainers.

Every frame, on purpose.

Every decision is written down. Every milestone lands in the repository.

Follow on GitHub