mcut

Local bridge

Connect MCP clients and local scripts to a live browser editor tab.

The local bridge keeps the browser editor tab as the source of truth while agents or scripts edit through MCP-style tools.

Live MCP server

bunx -p @mcut/mcp-server mcut-mcp-live --port 54319 --editor-url http://localhost:3000/editor

This starts a stdio MCP server and a localhost WebSocket bridge. It prints an editor URL with bridge parameters. Open that URL so the browser tab connects.

Bridge server

bunx -p @mcut/mcp-server mcut-bridge start --port 44737 --editor-url http://localhost:3000/editor

The bridge server exposes local HTTP RPC and WebSocket endpoints for a connected editor tab. Keep the process running while agents or scripts edit.

Bridge MCP adapter

bunx -p @mcut/mcp-server mcut-bridge mcp --port 44737

This exposes the running bridge as an MCP stdio server.

Utility commands

CommandPurpose
statusCheck whether the bridge is running and connected.
urlPrint the editor URL for a bridge port.
get-summaryPrint the live project summary.
get-projectPrint the live project JSON.
get-media-contextPrint live media and selection context.
get-transcriptPrint caption-derived transcript data.
search-transcriptSearch transcript text.
ensure-transcriptTrigger local browser transcription when available.
list-actionsList live browser editor actions.
actionRun a browser action by id.
dispatchDispatch a raw command.
operatorRun an editor operator by id.
undoUndo in the live editor.
redoRedo in the live editor.

When to use the bridge

Use the bridge when an edit needs browser-only capabilities, live editor state, local Whisper transcription, or exact parity with a running editor tab. Use the stdio project server when a project JSON file is enough.

For agent editing, use the high-level live actions before raw commands:

mcut-bridge action transcript.remove-silence --json '{"elementId":"e-video","paddingMs":120}'
mcut-bridge action effects.fade-open-close --json '{"elementId":"e-video","durationMs":500}'

transcript.remove-silence requires word-timed captions. Call ensure-transcript first when captions are missing. Do not use ffmpeg as a fallback for transcription or silence detection.

On this page