Can Claude Code read a video? The four ways, shown

Can Claude Code read a video? Here are the four ways, shown. You recorded the bug, but as of October 2026 Claude Code reads text and images, not video. Native video input is still an open feature request. So every method turns the video into something it can read. An agent needs three things: what you said, what you pointed at, and when it happened. Way one, paste a screenshot. Fine for a static glitch, but it loses your voice, the order of events, and everything between frames. Way two, paste the transcript. You keep your words in order, but not what you pointed at, so the agent has to guess which one you meant. Way three, a local skill runs ffmpeg and a speech model, and lines frames up with the transcript. It loses intent, and you own the setup. Way four, an MCP server serves the recording as data: a timed transcript, every mark you drew with its second, and an image of each mark on screen. Side by side, only the fourth way keeps your words, the order, and what you pointed at, with nothing to run on your machine. Here Claude Code reads each sentence beside its circle, finds the file and line for every ask, and files one issue. Pick a screenshot for one glitch, a local skill to stay local, or marks over MCP when the bug is where you point.