How to turn a screen recording into a prompt for Claude Code or Cursor

You cannot drop a screen recording into Claude Code or Cursor. Both take text and images, and neither one's documentation mentions video, so a recording has to become words, screenshots or both before an AI coding tool can use it. You can do that by hand: write down the steps, then paste a screenshot. Or you can let Lipo, our Mac app, do it. Its Visual Brief feature records your screen while you explain the problem out loud, then writes a structured prompt and pastes it into the chat you were using.

Can Claude Code or Cursor watch a screen recording?

Not directly. Anthropic's Claude Code docs describe three ways to add an image: drag and drop it into the window, paste it with Ctrl+V, or give Claude a file path. They do not mention video. Cursor's docs say you can attach images to the chat by dragging them in or pasting them, and dictate a prompt with the microphone in the chat box. They do not mention video either.

That leaves two kinds of input: words and still images. A recording helps the AI only after it has been turned into those.

How do you describe a bug so an AI can fix it?

Give the AI what a good bug report gives a person: where it happens, what you did, what you expected, and what happened instead.

• Where: the app, the page or screen, and the part of it. "The Settings modal in the web app, the Save button at the bottom."
• What you did: the steps, in order, starting from a state anyone can reach.
• What you expected, and what happened: "The modal should close and show a saved message. It stays open, and nothing is saved."
• What you already tried, so the AI does not suggest it again.
• A screenshot of the moment it goes wrong.

Use the real name of each thing, never "this" or "here". The AI cannot see your screen, so "the button on the right" is a guess, while "the Save button in the Settings modal" is a fact it can search the code for.

When is a recording better than a written report?

A recording wins when the problem is about movement or order. A layout that jumps after the page loads, a menu that opens in the wrong place, or a state that appears only after three clicks takes ten seconds to show and several minutes to describe. Design changes are similar. Pointing at a heading and saying "make this fit on two lines" is faster than writing out which heading you mean, and you are less likely to leave out the detail that matters.

How does Visual Brief turn a recording into a prompt?

1. Click into the chat or prompt box where the result should go, such as Cursor's chat or the Claude Code prompt in your terminal.
2. Start a Visual Brief from the Lipo icon in the menu bar, and choose full screen or a single window.
3. Talk through the problem while you show it. Point at things with your cursor, or mark them with the highlighter.
4. Stop when you are done. A recording can run for up to 5 minutes.

Lipo then transcribes what you said, matches each sentence to what was on the screen at that moment, writes the prompt, and pastes it where your cursor was. When a screenshot would help, it attaches one. The first time, macOS asks for the Screen Recording permission, and the Visual Brief docs explain the settings.

What does the finished prompt look like?

It starts with a short line that names the app and the screen, then gives one bullet for each change you asked for. When you showed several screens, the bullets are grouped under the name of each screen. Three rules shape what goes into it:

• Pointing words become names. If you say "remove this" while pointing at the FAQ link, the prompt says "Remove the FAQ item from the top navigation". The AI reading it cannot see where you pointed, so Lipo names the target from what was on screen.
• Your reasons stay. If you said "it looks cluttered", that stays as a short reason next to the change it explains, because it helps the AI make the right call.
• Corrections win, and cancelled requests disappear. If you say "make it shorter, no, keep the text but fit it on two lines", only the second version reaches the prompt. If you say "actually that's fine, leave it", the request is dropped completely.

Lipo does not add requirements you did not ask for. When it cannot read a target clearly, the prompt says so and asks the AI to check with you before changing it.

Can you explain it in your own language?

Yes. You can talk in any of Lipo's 50+ languages while you record, and Visual Brief matches what you said to what was on the screen, so a Hindi sentence about an English button still finds the button. Words that appear on the screen, such as button labels and page titles, are spelled the way the screen shows them, even when the speech recognition heard them differently. The guide to voice input in Claude Code and Cursor covers dictating ordinary prompts in your language.

What happens to the recording?

Recordings stay on your Mac for up to 7 days, together with the audio, the transcript, any screenshots and the prompt, and then Lipo removes them. Only what the analysis needs is sent for processing. Until then, each recording is listed in the Visual Brief tab, where you can copy the prompt again, ask Lipo to rewrite it from the same recording, rewatch the video, or delete it.

Treat a recording like a screenshot you are about to share. If an API key or a customer's details are on the screen, they are part of what gets analyzed, so hide them before you start.

Is Visual Brief free?

Visual Brief is included in every Lipo plan and in the 7-day free trial, which needs no credit card. Plans are on the pricing page. Visual Brief is labelled Beta, which means a few details may still change.

Frequently asked questions

Can Claude Code read a video file?

Its documentation describes adding images by drag and drop, Ctrl+V or a file path, and does not mention video. Turn the recording into a written prompt and screenshots first, by hand or with Visual Brief.

Can Cursor watch a screen recording?

Cursor's docs describe attaching images and dictating with the microphone, not video. Paste screenshots, or paste a prompt that Visual Brief wrote from the recording.

How long can a Visual Brief recording be?

Up to 5 minutes, of your full screen or of a single window.

Where does the prompt go when the recording ends?

Lipo pastes it into the app where your cursor was when you started. It is also saved in the Visual Brief tab, so you can copy it again later.

Does Visual Brief work with ChatGPT and Claude in a browser?

Yes. Lipo pastes the prompt into any text box, so it works in ChatGPT, Claude, Cursor, Claude Code in a terminal, and other AI tools.

What permission does Visual Brief need?

Screen Recording. macOS asks for it the first time you record, and you approve it in System Settings.