What your agent receives from a recording
Can Claude Code watch a video? It cannot play a video file, so Vidmatic turns your recording into text and images your agent can read, and serves them over MCP. The recording to issue skill is built on the same data.
Can Claude Code watch a video?
Not directly. It reads the words, marks and images listed below, all from one tool.
One call: get_recording_intel
Your agent calls get_recording_intel with the recording id and gets:
- The transcript, in short timed segments, each with its start and end time.
- Every mark you drew, each with the moment you confirmed it and, if you were talking, the sentence you were saying then.
- A thumbnail of the recording.
For each mark, get_annotation_image returns the image: the screen with the mark on it, so the agent sees the exact button you circled.
If your words and your mark disagree, your agent should trust the mark.
When the transcript is not ready
The response always says what state the transcript is in.
| State | What it means | What the agent does |
|---|---|---|
| Ready | The transcript is complete | Reads it |
| Processing | Uploading or being transcribed | Waits the number of seconds the response gives, then asks again |
| No audio | A timelapse, or a recording with no audio track | Works from the marks and the thumbnail |
| Failed | The audio could not be transcribed | Works from the marks and the thumbnail |
A processing transcript does not mean the recording is silent.
Recordings someone shared with you
A share link looks like /t/<id>?token=…. The Give this to your agent panel on that page adds --token to the prompt for you. Without the token your agent can read only your own recordings.
Reading it in your own words
You can also ask your agent in plain words: "Read Vidmatic recording <id> and list what I asked for." It can call the same tools.
Questions
- How does Claude Code read a screen recording?
- Claude Code cannot play a video file. Through the Vidmatic MCP server it reads a recording as a timed transcript, your on-screen marks with the second each was drawn, and images of what you circled.