egami-video for AI assistants

egami-video mcp serves egami-video's commands as tools to an AI assistant (Claude, GitHub Copilot, Cursor) over the Model Context Protocol, on standard input and output. It runs on the same server code as egami mcp: one tool per command (ffmpeg and ffprobe mode aside), with the command's options as arguments.

{
  "mcpServers": {
    "egami-video": { "command": "egami-video", "args": ["mcp"] }
  }
}

• Tools read and write only inside the folders the assistant opened, and those given with --allow <folder>. Without either they use the working folder.
• Outputs never replace inputs unless a call sets in_place. Standard input and output carry the protocol, so - is refused.
• Every result is an object: exitCode, written (the files written), output (the command's JSON or text) and messages. info answers in its JSON.
• Long jobs send progress notifications as they go, with the same line the terminal shows, and stop when the call is cancelled, leaving no partial file.
• A call may ask for a picture of the first file it wrote (preview): an image, or a video's first frame, fit to 1024 pixels.
• --max-inputs (1000 by default) bounds the files one call takes; --limit sets the safety limits for decoded frames.