What Is OpenChatCut? An Agent-Native AI Video Editor

OpenChatCut is an open-source, local-first AI video editor built for agents. Instead of asking an AI to generate a finished clip that you cannot easily revise, you give Codex, Claude, or the built-in agent an editing goal. The agent works on a real multitrack timeline, and every result stays visible and editable.

That distinction matters. OpenChatCut is not only a prompt box attached to a renderer. It is a video editor with clips, tracks, captions, effects, transitions, keyframes, audio, and project history—and an agent can operate those same editing primitives through MCP.

From a prompt to a real timeline

You can describe the outcome in ordinary language:

Turn this interview into a 45-second vertical cut. Remove long pauses, add readable captions, keep the speaker centered, and lower the music under dialogue.

The agent can inspect the project, propose changes, and perform the edit. When it finishes, the timeline is still yours. You can trim a cut by hand, adjust a caption, replace a transition, change an effect, or ask the agent for another pass.

The editor includes the familiar controls expected from a production timeline:

• Multiple video and audio tracks
• Move, trim, split, ripple edit, snapping, and markers
• Undo and redo history
• Keyframed properties and clip-level effects
• Captions, subtitle import and export, and translation workflows
• Music, sound effects, voiceover, loudness tools, and auto-ducking
• Motion graphics, LUTs, shaders, zooms, and transitions

Why MCP is part of the design

Model Context Protocol gives external agents a structured way to read and modify an OpenChatCut project. It means the same project can be operated by the built-in agent, Codex, Claude Code, or another MCP-compatible client without inventing a separate automation format for each one.

OpenChatCut exposes a local Streamable HTTP MCP endpoint:

http://localhost:5199/api/external-mcp/mcp

The tools are project-aware. An agent can inspect media and tracks, add or edit timeline items, manage captions and transitions, and export the result. Actions use the same project model as the editor UI, so automated edits remain understandable instead of becoming an opaque generated file.

Local-first by default

Video projects contain large files and often include private footage. OpenChatCut keeps projects and media local by default. Project data is stored under ~/.openchatcut, while the browser editor uses IndexedDB as a local cache. Optional remote storage can be configured when a workflow needs it.

Local-first does not mean isolated. Online media search, AI generation providers, external agents, and remote storage can be connected when useful. The difference is that the editing project does not require a hosted account to exist.

Production output, not a dead end

An agent-assisted edit should not lock you into one application. OpenChatCut can export finished video, audio, captions, editable project archives, and FCPXML for continued work in compatible professional editors. Project history and import/export make it possible to move, back up, and revisit an edit.

Who is OpenChatCut for?

OpenChatCut is useful when you want AI assistance without giving up timeline control:

• Creators turning long recordings into short-form clips
• Developers building agent-driven media workflows
• Teams that need repeatable editing actions
• Editors experimenting with AI while keeping manual control
• Researchers testing MCP and multimodal agent interfaces

The project is under active development. The current recommended path is to run it from source, connect your preferred agent, and keep the final editorial decision in the timeline.