Cursor

How to let Cursor watch a video

Click Add to Cursor and it writes the MCP server into Cursor's settings. From then on the agent can call watch_video on any link or on a file in the workspace.

Free tier: 60 credits a month, no card. One credit is about a minute of video.

What this changes

The reason this is useful inside an editor is bug reports. Somebody sends a 40-second screen recording of something going wrong; the agent watches it, finds the frame where the UI breaks, reads the error text off the screen, and starts from the line of code that produced it. Without something that can see the recording, that whole chain begins with you writing a description of what you just watched.

Setting it up

  1. 01

    Click the install button

    It opens Cursor through a deeplink and adds the server to your MCP configuration. If nothing happens, Cursor is not installed — paste the URL into Settings → MCP instead.

  2. 02

    Sign in

    The first tool call opens a browser to approve the connection. It is once per machine.

  3. 03

    Drop a recording into the chat

    A path in the workspace or a URL. Ask what happens, and at what second.

Things worth asking it

Each of these produces an answer with a second attached, because the timestamp is burned into the frame the model is looking at — it cannot lose track of when something happened.

  • Watch ./bugs/report-441.mov — at what second does the list stop rendering?
  • Read the error message visible in this screen recording.
  • This design reference clip — describe the transition between the two states, frame by frame.

How closely it looks is a choice

Six levels, picked by the agent from the question. Surveying an hour and reading a single frame of animation are different jobs, and charging the same for both would make one of them pointless.

LevelRateAnswers
overview0.5–2 fpsWhat is this video, start to end
standard5 fpsWhat happens when — cuts, overlays, entrances
detail5 fpsObjects and text blocks, tiles four times larger
fine10 fpsType, logos, subtle transitions
motion30 fpsAnimation, easing, dropped frames
singleone frameReading text at full resolution

Questions

Does the deeplink work if Cursor is not open?
Yes, the OS launches it. If the click does nothing at all, Cursor is not installed on this machine — the server URL is shown on the install page for pasting in by hand.
Can it read text in a screen recording?
That is what inspect_frame is for: one frame at 1560 pixels wide, which is the size at which small UI text stays legible to a model. watch_video finds the moment; inspect_frame reads it.

Or somewhere else

One account, every client. The connection is attached to you rather than to a machine, so adding a second one costs nothing.

Stop describing the video to your agent.