An AI agent connected over MCP now sees frames of a screen recording, not only the transcript of what the reporter said. The new tool get_recording_frames returns a still image for each of up to six moments the agent picks, such as the end of a transcript line or the moment a request failed. It works on any screen recording, with or without a transcript, and a call takes a few seconds while BugDump downloads the recording.
get_report sent every attachment twice, its transcript included, so a narrated recording took twice the room in the agent's context. It now sends each attachment once.
See Using BugDump with Claude Code (MCP), and An AI Agent Can't Watch a Bug Recording, So It Gets the Frames for how it was built.