Skip to content
All features

Screenshots of Modals, Dropdowns and Tabs

Automatically capture the open modal, the expanded dropdown and each tab of your web app for your documentation, and recapture them when your UI changes. Describe the clicks in plain words. No Playwright scripts to write or maintain.

The problem

Docs need the settings modal open, the menu expanded and the second tab selected, not just the page as it loads. Teams either retake these by hand after every release or write browser scripts that click through each state, and those scripts break whenever a button is renamed or the page's markup changes.

How ScreenshotAgent solves it

Tell ScreenshotAgent which page to open and what to click, in plain words ("Click Settings, then the Billing tab"). The capture agent opens the page, does the steps and takes the screenshot, and follows the same steps on every recapture. In the app, each state is a view nested under its page: describe it, click on the screenshot to record the clicks, or let Auto-Discover explore the page and create views for the states it finds. Point & Click views and views found by Auto-Discover record the clicks that open them, so when they sit directly on a page their recaptures replay those clicks and fall back to the AI agent when a click no longer works.

Example: an agent captures the Billing tab of the settings modal

Connect the ScreenshotAgent MCP server to Claude Code with a team API token that has the read, captures and manage scopes. With the app's login saved on a project, ask:

A docs page that embeds that image URL shows the current Billing tab once each recapture goes live, with no new upload. Claude Code is told never to ask for your password: the login is saved in the app. In ChatGPT, the ScreenshotAgent plugin can take the screenshot the same way; set its schedule and make it public in the dashboard.

Get ScreenshotAgent in ChatGPT
Claude Code

> Take a screenshot of our dashboard with the Settings modal open on the Billing tab, keep it updated weekly, and give me the image URL for our docs.

  1. 1.The assistant calls create_screenshot with the dashboard URL, the project and the instructions "Click Settings in the top right, then the Billing tab."
  2. 2.The capture agent opens the dashboard, signs in with the project's saved login if needed, clicks Settings, waits for the modal, clicks the Billing tab and saves the screenshot.
  3. 3.The assistant calls keep_current with a weekly schedule and public: true, which returns the screenshot's image URL. Making a screenshot public needs a team member who can approve screenshots.
  4. 4.Every weekly recapture follows the same instructions. With a saved login the agent follows them each time, so after a redesign it finds the buttons by their description. Turn on Needs approval to check each new version before it goes live.

Expected outcomes

  • Every state is its own screenshot with its own permanent embed URL and version history, and follows its page's approval setting.
  • Views nest: a modal on a page, then a tab inside that modal.
  • States behind a login are captured with the project's saved login.
  • Views you describe in plain words are followed by the AI agent on every recapture, so a renamed or moved button is still found by its description.
  • Works from the app, the MCP server and the ScreenshotAgent plugin in ChatGPT, so AI assistants and coding agents can capture states too.

FAQ

Do I need to write Playwright or Selenium scripts?
No. Write the steps in plain words, and the capture agent follows them on every recapture. Or use Point & Click to click on the screenshot where the agent should click: the clicks are recorded on the first capture and, for views directly on a page, replayed after that.
How would I do this with Playwright instead?
Write a test that opens the page, signs in, clicks what opens each state (for example page.getByRole('button', { name: 'Settings' }).click()), waits for it to appear and calls page.screenshot(). Run it in CI after each deploy and upload the images to your docs. That suits teams whose engineers own the docs. The upkeep is the locators that break when a button is renamed or the markup they rely on changes, the login step, and storing and hosting the images. ScreenshotAgent takes over those parts: plain-language steps, a saved login, and permanent image URLs.
What about Storybook and Chromatic?
Storybook is a workshop for building UI components and pages in isolation, and Chromatic takes visual snapshots of your Storybook stories or Playwright and Cypress tests. Both run from your codebase. Help docs need a picture of your running app, signed in, kept current at an image URL the article can embed, and that is what ScreenshotAgent captures.
What happens when a button is renamed or moved?
Views you describe in plain words are followed by the AI agent, which reads the page each time, so the button is found by its description. Point & Click views and views found by Auto-Discover replay their recorded clicks first when they sit directly on a page; when a click no longer works, the agent takes over and that new version waits for review before it goes live.
Can it find the states on a page for me?
Yes. Click Auto-Discover in a screenshot's view tree and the agent explores the page, finds elements that open new states (modals, menus, tabs) and creates a view for each.
Does it work for pages behind a login?
Yes. Save the login on the project or page and views use it. Credentials are encrypted at rest and kept out of logs. A view that replays its recorded clicks signs in again when the saved session has expired; views the AI agent follows are given the login and sign in as part of the capture.

Related features

Automate this workflow in your team

Start a free trial and connect your first project to see ScreenshotAgent capture and keep pages fresh.

Start free trial14-day trial · 5 captures free · No card required