How to give Cursor a screen recording (November 2026)
Checked and updated
Cursor's agent can't play a video, so give it the recording as a change list: add Tellscreen's MCP server to Cursor once, then hand the agent any video file or link. It gets every change you asked for, with your exact words and a screenshot of each moment.
Add the server (once)
Put this in ~/.cursor/mcp.json for every project, or .cursor/mcp.json in one project:
{
"mcpServers": {
"tellscreen": {
"url": "https://tellscreen.com/mcp"
}
}
}
There's nothing to sign in to. Cursor shows four tools: convert_local_video, convert_video_link, record_screen and get_change_list.
Then, in the agent chat
- A video file: "Make the changes in ~/Desktop/review.mov." The agent asks Tellscreen how to send it and runs one
curlcommand that uploads it and prints the change list. - A link: "Here's the recording: https://…/review.mp4". It works with a direct video link or a public Google Drive or Dropbox link. Loom pages can't be fetched; download the Loom video first.
- Nothing recorded yet: "Let me show you." The agent gives you a recording link; you record your screen and talk, press Stop, and the change list comes back to the agent.
The video is uploaded as a private video, and the answer includes a link to it in your Tellscreen library.
Without MCP
Run this in Cursor's terminal, or ask the agent to:
curl -fsS -T review.mov "https://tellscreen.com/api/v1/convert?wait=110"
Why not just attach the video?
The models behind Cursor read text and images. A screen recording's requests are mostly spoken, and without the sound no model we tested could name a single change. A frame-by-frame dump is also expensive: a 10-minute walkthrough at one frame every 2 seconds is about 400,000 tokens.
What it costs
Converting a video counts as a video and a change list on your Tellscreen plan: Free includes 5 change lists a month, Pro ($8 a month) 200. See pricing.
Sources
- Cursor docs, Model Context Protocol: the
mcp.jsonformat and file locations, read 10 November 2026