What is video-mcp?

video-mcp is a server that plugs into AI coding tools (using the Model Context Protocol, or MCP, a standard interface for giving AI assistants new capabilities) and lets the AI agent record your screen, analyze what happened, and produce a narrated video walkthrough. The main use case is helping an AI agent document a bug or demonstrate a workflow without any manual video editing. The tool covers the full pipeline from recording to a finished video. It can start and stop a screen recording, then automatically detect the key moments where the interface changed, cut out relevant clips, zoom into specific areas, and slow down or speed up sections. Once the visuals are assembled, it can generate a spoken voiceover from a script using text-to-speech (either online or offline with a downloaded voice model), overlay the narration, add background music, burn text labels onto frames, animate cursor highlights, and stitch everything together with transitions or hard cuts. The project currently runs on Linux with the X11 display system only. It depends on Node.js, Python, ffmpeg, and tesseract-ocr for detecting on-screen text elements. It is at an alpha stage and was tested on a single Ubuntu machine. The non-recording tools for analysis, annotation, narration, and assembly are described as mostly platform-agnostic and may work on other systems with minor changes. The full README is longer than what was provided.