I built an open-source video editor where the AI actually drives the timeline

As developers and technical creators, we’ve all experienced context-switching fatigue. When you need to spin up a quick product demo, short, or vlog, you're usually forced to jump out of your development environment, spin up heavy, closed-source cloud SaaS tools, and manually scrub timelines just to cut a few clips.

I wanted an editor where the AI doesn't just output text recommendations or code snippets, but has direct tool-calling capabilities over the actual timeline.
That's why I built SynthCut—a local-first, open-source video editor designed to be fully driven by an AI agent.

Under the hood, it exposes functionality through Model Context Protocol (MCP) servers. This means you can drop it right into your existing agentic workflow. It hooks seamlessly into tools like Claude Code, Cursor, and Windsurf, exposing the video environment as a native toolset for your assistant.

Instead of dragging UI elements around, you can just write a prompt directly in your terminal or IDE:
"Use /ai-video-editor, and all the clips present here to prepare a mini vlog video for the social media post."

The agent handles the rest programmatically: parsing asset metadata, analyzing content bounds, sequencing hooks, syncing layout coordinates, and rendering frames without requiring manual timeline micromanagement.

I deliberately built this to be 100% local and open-source. Your raw video assets stay on disk, there are no cloud-rendering bottlenecks or forced subscriptions, and you retain total control over the execution layer.

You can check out the source code and documentation on the SynthCut GitHub Repository(https://github.com/Relo-video/SynthCut). I’d love for fellow devs to check it out, test the MCP integration, and let me know how it handles your local workflows or where the rough edges are.