nhatvu148/video-transcriber-mcp ? reverse-engineered prompt

Build me a Model Context Protocol server that can take a video link or a local video file and turn it into a transcript.

I want it to work with a lot of sites like YouTube, Vimeo, TikTok, X, Facebook, Instagram, Twitch, and other places yt dlp can handle, plus local mp4, mov, mkv, and similar files. It should download the audio, transcribe it with whisper cpp, and let me choose the model size, with auto language detection when possible. Please also support saving the transcript as plain text, JSON, and Markdown.

Make it easy to use from Claude Code or any MCP client, and include helpful tools for listing supported sites, checking past transcripts, getting the latest one, deleting old ones, and cleaning up everything if needed. If the user wants, it should also be able to send transcription work to a remote worker instead of doing it locally. Keep it reliable, with clear errors and retries, and make setup instructions simple for macOS, Linux, and Windows.