YouTube Vision acts as a smart pair of eyes for AI assistants, allowing them to "watch" and understand YouTube videos directly. Instead of just relying on text transcripts, this tool uses visual analysis to help AI models see what is happening on screen. It makes it possible to get clear descriptions of video content, ask specific questions about what was shown, or get a quick summary of a long presentation without having to watch the entire thing. The technical core of this MCP server is built on the Google Gemini Vision API, which provides the reasoning capabilities needed to interpret video frames. The server includes specialized tools for different tasks: `summarize_youtube_video` can generate overviews of varying lengths, while `extract_key_moments` identifies critical timestamps and provides descriptions for each. It also features a flexible Q&A tool, `ask_about_youtube_video`, which allows for deep dives into specific visual details or general content inquiries. For developers and researchers building AI-driven workflows, this server bridges the gap between Large Language Models and rich multimedia data. It supports configurable model selection, such as `gemini-1.5-flash` or `gemini-2.0-flash`, via environment variables to balance performance and cost. By following the Model Context Protocol (MCP) and communicating through standard input/output (stdio), it offers a plug-and-play solution for giving AI agents the ability to programmatically parse, analyze, and retrieve information from the world’s largest video platform.
Category: Design, Media & Creative
Tags: computer-vision, gemini, multimedia, video-analysis, youtube
bash npx -y @smithery/cli install @minbang930/youtube-vision-mcp --client claude Option 2: Manual Installation (from Source) 1. Clone the repository: bash git clone https://github.com/minbang930/Youtube-Vision-MCP.git cd youtube-vision 2. Install dependencies: bash npm install 3. Build the project: bash npm run build ---json { "mcpServers": { "youtube-vision": { "command": "npx", "args": [ "-y", "youtube-vision" ], "env": { "GEMINI_API_KEY": "YOUR_GEMINI_API_KEY", "GEMINI_MODEL_NAME": "gemini-2.0-flash" } } } } Environment Variables: * GEMINI_API_KEY (Required): Your Google Gemini API key. * GEMINI_MODEL_NAME (Optional): The Gemini model to use (e.g., gemini-1.5-flash). Defaults to gemini-2.0-flash. ---ask_about_youtube_video: Answers a specific question about a video or provides a general description. * youtube_url (string, required): The URL of the YouTube video. * question (string, optional): Specific question about the video. * summarize_youtube_video: Generates a summary of the video. * youtube_url (string, required): The URL of the YouTube video. * summary_length (string, optional): 'short', 'medium', or 'long'. Defaults to 'medium'. * extract_key_moments: Extracts timestamps and descriptions of key moments. * youtube_url (string, required): The URL of the YouTube video. * number_of_moments (integer, optional): Number of moments to extract. Defaults to 3. * list_supported_models: Lists available Gemini models that support content generation. * No input parameters. ---extract_key_moments and summarize_youtube_video tools. Because it uses Gemini Vision, it can identify topics based on slide changes and visual cues, not just audio transcripts. Example: A medical student provides a 2-hour lecture on "Neurological Disorders" and asks the MCP to "Extract key moments where the professor discusses Parkinson's disease symptoms and summarize the visual diagrams shown during those segments."ask_about_youtube_video tool, developers can query the visual state of the video. The Gemini Vision API can "see" the code on the screen and describe specific configurations or UI steps. Example: A developer stuck on a Cloud deployment tutorial asks: "What were the exact values entered into the 'Environment Variables' screen at the 8:45 mark of this video?" The MCP analyzes the frames and returns the text visible in the video's UI.summarize_youtube_video and extract_key_moments, creators can identify patterns in how successful videos are visually paced and structured. Example: A marketing analyst provides URLs for the top 5 trending videos in a specific niche and asks: "Summarize the first 30 seconds of each video and describe the common visual elements used to hook the audience in these clips."Part of MCP Servers