The work happens at Gemini, so the key and the model named in GEMINI_MODEL_NAME decide both what the answers cost you and how good they are. The tool list is read-only over a URL: nothing clips, downloads or exports a transcript, and the output is always text about the video. extract_key_moments is the one that changes a workflow — timestamps with descriptions give you somewhere to jump to instead of a paragraph to trust.
A Gemini Vision wrapper for YouTube URLs: summary, question answering and key-moment extraction over a video you name, returned as text.
- summarize_youtube_video summarises a video from its URL, at a summary_length you choose.
- ask_about_youtube_video answers a specific question about the video, and returns a general description when no question is passed.
- extract_key_moments returns key moments as timestamps with descriptions, as many as number_of_moments asks for.
- list_supported_models lists the Gemini models that support the generateContent method, which is what the other tools call.
A Gemini API key in GEMINI_API_KEY and a model name in GEMINI_MODEL_NAME — list_supported_models returns the models that support generateContent, which is what summarize_youtube_video, ask_about_youtube_video and extract_key_moments call.
One command plus a key — npx -y @smithery/cli install @minbang930/youtube-vision-mcp --client claude, then supply credentials