EffectCraft MCP: a compositor your agent can drive

Two ways to run it

Set up with Claude Code

The release packages ship effectcraft-cli next to the app:

Make a working folder, then register the server, confined to that folder:

mkdir -p "$PWD/effectcraft-work"
claude mcp add effectcraft -- effectcraft-cli mcp --autosave \
  --automation-read-root "$PWD/effectcraft-work" \
  --automation-write-root "$PWD/effectcraft-work"

Or add it to .mcp.json in your project:

{
  "mcpServers": {
    "effectcraft": {
      "command": "/path/to/effectcraft-cli",
      "args": ["mcp", "--demo",
        "--automation-read-root", "/absolute/path/to/workspace",
        "--automation-write-root", "/absolute/path/to/workspace"]
    }
  }
}

--demo starts with the demo project and --project file.ecproj with yours. To watch the agent work in the desktop app, start effectcraft --control 9877 and use ["mcp", "--bridge", "9877"]. The control port has no authentication, so only enable it while you use it.

What the agent gets

• Commands: list_commands and describe_command to discover ids and their parameters, execute_command to run one, batch for several in one undo step.
• Project and properties: get_state, get_project, get_comp, get_layer, get_property, set_property (values and expressions) and add_keyframe with easing.
• Effects and fonts: list_effects, add_effect with its settings in one call, list_fonts.
• Seeing the result: render_frame returns a PNG the model can look at, with alpha if you ask; in bridge mode screenshot and the ui_* tools see the window.
• Scripts and history: run_script runs JavaScript with an After Effects-style object model; undo, redo and history walk the undo tree.

A typical session

1. execute_command comp.new with a name, 1920 × 1080, 30 fps and a duration.
2. execute_command layer.newText with the text, size and fill; the reply names the new layer.
3. add_keyframe on transform/position with two keys and easyEase.
4. add_effect to apply Gaussian Blur or Glow with its values.
5. render_frame to check a frame, then save_project, or renderQueue.add and renderQueue.render for the video.

Drive the browser build with window.effectcraft

The web build on this site exposes the desktop control channel as promises on its app page. Open effectcraft.run/app/ in a browser your agent controls (or reach the editor’s iframe with document.querySelector('iframe').contentWindow.effectcraft), then:

await effectcraft.execute("layer.newSolid", {color: "#ff8800"});
const frame = await effectcraft.renderFrame({max_side: 640});  // {width, height, png: base64}
await effectcraft.addFile(blob, "logo.png");                   // store and import into the Project panel
await effectcraft.request("ui.menu.invoke", {id: "renderQueue.add", params: {format: "gif", resolution: 0.25}});
await effectcraft.execute("renderQueue.render", {wait: false}); // renders in a worker; the GIF downloads
await effectcraft.saveToBrowser("/my-project.ecproj");          // File › Open Recent lists it after a reload

Engine commands take the same ids and parameters as MCP and the CLI. effectcraft.commands() lists them, effectcraft.info() reports the graphics backend, storage and workers in use.

Without MCP: the CLI on its own

# Set a property and save, in one call
effectcraft-cli set Main '#1' transform/position '[100,360]' --time 0 main.ecproj --save

# Render a comp headless
effectcraft-cli render --comp Main --out main.mp4

# Export Lottie
effectcraft-cli exec file.exportLottie '{"comp":"Main","path":"main.json"}' main.ecproj

FAQ

Source: EffectCraft’s agents guide, MCP conventions and web build notes. Flags can change between releases; check effectcraft-cli mcp --help.