FilmCraft MCP: a video editor your agent can drive

Two ways to run it

Set up with Claude Code

The release packages ship filmcraft-cli with the app:

# Linux, or macOS with the CLI unzipped onto PATH
claude mcp add filmcraft -- filmcraft-cli mcp

# Windows, default install folder
claude mcp add filmcraft -- "C:\Program Files\FilmCraft\filmcraft-cli.exe" mcp

Or add it to your client’s configuration, here as a bridge to the running app:

{
  "mcpServers": {
    "filmcraft": { "command": "/abs/path/filmcraft-cli", "args": ["mcp", "--bridge", "127.0.0.1:9876"] }
  }
}

The control port listens on 127.0.0.1 only but has no authentication, so only enable it while you use it.

What the agent gets

• Commands: command_list (id, label, menu, shortcut, parameters, enabled now), command_run for one command, command_batch for several in order. Edits are undoable.
• State: project_inspect for bins and items, sequence_inspect for tracks, clips, effects, transitions, markers and the playhead, or doc_inspect for both.
• Media: media_import imports files by absolute path.
• Seeing the result: render_frame returns a PNG of the Program frame at any time; in bridge mode ui_screenshot captures the window or one panel.
• The UI itself (bridge mode): ui_elements, ui_click, ui_drag, ui_key, ui_type and ui_control, with a stable automation id on every widget.

Read-only tools carry MCP’s readOnlyHint, so a client can let inspects and renders run without asking and confirm edits.

A typical session

1. project_inspect to find item ids, then media_import for new footage.
2. command_run file.newSequence with a name, size and frame rate.
3. command_run timeline.place for each clip, timeline.razor or sequence.addEdit to cut, edit.rippleDelete to close gaps.
4. command_run effects.setParam to grade a clip with Lumetri parameters, and render_frame to look at the result.
5. command_run file.exportMedia with a path, format: "h264" and wait: true.

The FilmCraft team built the trailer and grades in its README screenshots this way, with an agent driving the running app.

Drive the browser build with window.filmcraft

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

await filmcraft.importFiles(input.files);                          // File / FileList / File[]
await filmcraft.execute("file.newSequence", {name: "Cut 1"});
await filmcraft.execute("timeline.place", {item: 12, track: "V1", seconds: 0});
await filmcraft.request("ui.playback", {action: "play"});
const {pngBase64} = await filmcraft.screenshot();                  // the canvas as PNG
const {job, path} = await filmcraft.execute("file.exportMedia", {format: "h264", path: "/exports/out.mp4"});
const bytes = await filmcraft.readFile(path);                      // once jobs.list says it finished

Engine commands take the same ids and parameters as MCP and the CLI. filmcraft.commands() lists them, filmcraft.files() lists the browser’s virtual files, and filmcraft.info() reports the graphics backend, compositor, storage and WebCodecs support. Commands that need threads on the desktop (render previews, proxies, Project Manager, mask tracking) report an error in the browser for now.

Without MCP: the CLI on its own

# List every command
filmcraft-cli commands

# Razor the active sequence at 3.5 s and save
filmcraft-cli --project cut.fcproj --save exec timeline.razor seconds=3.5

# Export the active sequence as H.264 and wait for it
filmcraft-cli --project cut.fcproj export out.mp4 --format h264

# One Program frame to PNG
filmcraft-cli --project cut.fcproj render --seconds 12 --out frame.png

FAQ

Source: FilmCraft’s agents guide, control protocol and web build notes. Options can change between releases; check filmcraft-cli help.