Because analysis is multi-modal and timed, a hit points at a span inside a video rather than at the video. That is what makes cut me something from the parts where this happens a request the server can act on, instead of a brief for a human editor.
A Python MCP server for Video Jungle. Videos are added from a URL and analysed multi-modally — audio and visual — so a search returns matches with metadata about what happens and when, which is enough to generate an edit directly.
- A video added from a URL, analysed, and returned as a vj:// reference
- Search across your library on embeddings and keywords, with in-video timing in the results
- An edit generated from a single video, or from a set of them
- An existing edit updated live — if Video Jungle is open, it changes as you go
- Projects created to hold generative scripts, analysed videos and images, and their assets listed
- An OpenTimelineIO project downloaded to your machine to open in DaVinci Resolve Studio
A Video Jungle account and its API key, added to your client config. edit-locally additionally needs DaVinci Resolve Studio already running before the call, since it hands the project to a live instance.
One command plus a key — uvx video-editor-mcp YOURAPIKEY, then supply credentials