sakshamjain2301/videocontext-mcp ? reverse-engineered prompt

Build me a local first tool that lets an AI assistant understand YouTube videos and videos on my computer without sending the whole file anywhere.

I want it to run as a small local server that works through stdio, with a simple command line tool for inspecting a video, preparing it, and then reading a timestamped transcript, a short timeline, and a few selected frames. It should support both YouTube links and local videos in a safe way, keep everything bounded and private, and let me pick only the parts I need instead of processing the whole video every time.

Please include a way to check the setup, install a Whisper model manually, and delete or expire jobs when I am done. If you need to look up current docs for any of the libraries or video tooling, go ahead and do that. Focus on making the workflow smooth, secure, and easy to use from an AI client.