How to Start a New Browser Recording in Claude Scope

Set up and launch a browser screen recording to capture UI interactions and generate structured system prompts for AI coding agents.

Starting a new recording is the first step in the Claude Scope pipeline. You fill in a few details about the session, grant screen-sharing permission, and Claude Scope captures your browser interaction as a video — ready for frame extraction and prompt generation.

Recordings are saved as WebM video files. Files larger than 100 MB will be rejected at upload. With the default auto-stop at 30 seconds, most recordings stay well under this limit. Longer recordings captured with auto-stop disabled may exceed it.

Recording form fields

Before you start recording, fill in the session details on the New Recording page.

Field	Required	Details
Recording title	Yes	A short label for the session, such as Checkout flow bug. Shown in the Sessions list.
App URL to inspect	Yes	The URL Claude Scope uses for its Playwright ARIA inspection pass. Must start with http:// or https://.
Notes	No	Optional context for the agent — describe what to focus on. Maximum 500 characters.
Auto-stop at 30 seconds	—	Toggle. On by default. Disable to record past 30 seconds.

Chrome and Edge are recommended browsers. Firefox has partial support for the Screen Capture API and may not expose all tabs or windows in the sharing picker.

Starting a recording

1. Open the New Recording page

   From your dashboard, click New Recording. The recording setup form opens.

2. Fill in the recording title

   Enter a descriptive title in the Recording title field. This field is required — the Start Recording button stays disabled until both the title and app URL are valid.

3. Enter the app URL to inspect

   Type or paste the URL of the page you want to capture and inspect. The URL must begin with http:// or https://. Claude Scope validates the format when you leave the field — an inline error appears if the URL is invalid.

4. Add optional notes

   Use the Notes field to give the agent additional context, such as which elements to focus on or what behavior to describe. The character counter tracks your input up to the 500-character limit.

5. Configure auto-stop

   The Auto-stop at 30 seconds toggle is on by default. Leave it on for most recordings. If you need to capture a longer interaction, disable the toggle — recording continues until you click Stop Recording manually.

6. Click Start Recording

   Click Start Recording. Your browser opens its screen-sharing permission dialog. Select the tab or window you want to share and confirm. Claude Scope cannot start recording until you grant permission.

   If you dismiss the dialog or deny permission, an error message appears and no recording is started.

7. Interact with your app

   Once permission is granted, the recording screen shows:

   • A pulsing red indicator and RECORDING label
   • A large timer counting elapsed seconds
   • A progress bar tracking time toward the auto-stop limit

   Navigate through the UI states you want Claude Scope to analyze. The progress bar fills as you approach 30 seconds.

8. Stop the recording

   Recording stops in one of two ways:

   • Auto-stop — at 30 seconds the recorder stops automatically (if the toggle was on).
   • Manual stop — click Stop Recording at any time.

   You can also stop sharing via the browser’s own screen-sharing controls; Claude Scope detects this and stops the recorder cleanly.

9. Wait for upload and processing

   After stopping, Claude Scope saves the video to local storage (IndexedDB), then automatically uploads it for processing. The Processing page opens and shows the pipeline progress — frame extraction, vision analysis, Playwright inspection, and prompt synthesis.

What happens after recording

Once the upload completes, Claude Scope runs the following pipeline stages automatically:

Frame extraction

Key frames are extracted from the video using SSIM-based frame differencing to identify meaningful UI state changes.

Vision analysis

Each extracted frame is analyzed by a Vision AI model to detect UI elements and describe interactions.

Playwright inspection

Claude Scope visits the app URL you provided and captures a full ARIA accessibility snapshot of the live page.

Prompt synthesis

The visual timeline and structural inspection are combined into a structured system prompt for your chosen AI coding agent.

When all stages complete, you are taken to the Frame Review page to inspect the extracted frames before generating the final output.