« All posts

SynthCut: An Open-Source Video Editor Built for AI Agents

SynthCut is an open-source, MCP-driven video editor where AI clients control local FFmpeg edits offline, with multi-track editing and OTIO support.

SynthCut is an open-source desktop video editor designed to be operated by AI clients rather than humans directly. It exposes itself as an MCP (Model Context Protocol) server, letting any MCP-compatible client — such as Claude Desktop — drive real, local FFmpeg-based edits fully offline. Users describe an edit in plain language, the AI executes it, and the human reviews and corrects the result.

The project provides a full multi-track, frame-based timeline editor with over 90 MCP tools covering import/cut/trim, per-clip transforms with keyframe animation, color grading, an effects stack, Whisper-based captioning, subject-aware auto-reframe, motion graphics, and media intelligence features like transcript and semantic visual search. An Electron+React UI stays live-synced with the core engine over WebSocket, so both the AI and the human operate on the same shared editing state.

Licensed under GPL-3.0, SynthCut supports OpenTimelineIO (OTIO) interchange, allowing timelines to move to and from professional tools like DaVinci Resolve. Windows installers are distributed with SHA256 checksums for verification, since the unsigned open-source build triggers a standard SmartScreen warning. Notably for engineers, every editing capability is defined once in a shared RPC registry, functioning simultaneously as both an AI-callable tool and a UI action.