Enable agents to view video links and read subtitles
Copy a video link from your browser and paste it into the chat window, and the agent selects key scenes from the video and transcribes the spoken content into text. There is no need to open video files directly or extract subtitle files separately; the agent automates the process of 'watching' and understanding the video. This tool is a skill installed and used in agent environments such as Claude Code or Codex.
Converting video evidence to text
The tool separates visual and audio information from videos and converts them into a format that agents can process. By default, it extracts frames from specific points in the video and converts subtitles to text if they exist. For videos without subtitles, it runs a local speech recognition model or calls a cloud API to generate them. These frames and subtitles are bundled with timestamps and passed to the agent. Based on this data, the agent summarizes the video content or describes scenes at specific moments.
Two processing engines
Users can choose between two processing methods. The first is the Gemini engine, which uses Google's AI models to analyze videos directly. This method can process YouTube links immediately, while local files are uploaded to Google servers before analysis. The second is the local engine, which uses FFmpeg and yt-dlp on the user's computer to extract frames and generate subtitles. Because the local engine does not upload videos to external servers, it is suitable for privacy-focused environments. The default setting automatically selects Gemini if a Gemini API key is present, otherwise it uses the local engine.
Installation and initial setup
It can be installed in various agent environments, including Claude Desktop, VS Code, the terminal, and Codex. For Claude Desktop users, add the marketplace from the 'Customize' menu and install the Watch skill. For Claude Code users in the terminal, enter the /plugin marketplace add bradautomates/claude-video command to install. Codex users can enter installation instructions in the message window or run terminal commands. After installation, Python 3.10 or higher, FFmpeg, and yt-dlp must be present on the system, and the agent automatically checks these dependencies and guides you through the installation process.
Pre-use checks
This tool works only in specific agent environments and cannot be used in Claude Chat or Cowork. When using the local engine, Python, FFmpeg, and yt-dlp must be installed with the latest versions, and a JavaScript runtime like Deno is required to process YouTube videos. Enabling speech recognition requires a minimum of 3GB of disk space and 8GB of RAM for a local WhisperX installation, which has been verified on Apple Silicon macOS. The cloud engine requires a GEMINI_API_KEY, and since local files are uploaded to Google servers, the local engine is recommended for sensitive videos. yt-dlp requires frequent updates due to site changes, so maintaining the latest version is important.
