How to beta test AgentScreenshots

Thanks for testing AgentScreenshots. The goal is simple: give your coding agent a fast way to see rendered UI without opening a browser, clicking around, or using a slow MCP workflow.

This guide covers the exact commands worth trying.

1. Install the CLI

AgentScreenshots is published on npm. Install it globally so every project and every agent can call the same command:

npm install -g agentscreenshots

Then open your AgentScreenshots dashboard, generate a key from your free, paid, or beta license, and authenticate the CLI:

agentshot auth ags_live_xxx

Run the doctor once:

agentshot doctor

doctor checks Node, browser launch, config, output paths, API reachability, and license status.

2. Test live page screenshots

Start with any public URL. This should feel boring:

agentshot "https://agentscreenshots.com" "home.png"

For pages with lazy-loaded content, animation, or infinite-scroll behavior, trigger scroll and wait before capture:

agentshot "https://agentscreenshots.com" "home-scroll.png" --scroll --wait 1000

Capture a vertical range when you only need part of a long page:

agentshot "https://agentscreenshots.com" "middle.png" --from 1400 --to 2400

Capture the top of the page:

agentshot "https://agentscreenshots.com" "hero.png" --height 1200

Capture mobile:

agentshot "https://agentscreenshots.com" "mobile.png" --viewport 390x844 --scroll --wait 1000

The thing to check: do the screenshots contain the content you expected, at the size you expected, without manual browser work?

3. Test localhost websites

This is the main workflow for AI coding agents.

Start your local frontend dev server, then capture the app directly:

agentshot "http://localhost:5173" "local-home.png" --scroll --wait 1000

No tunnel is needed. The browser runs on your machine, so localhost works naturally.

4. Test selector screenshots

Selector capture is the workflow to try when your agent is iterating on a specific section instead of a whole page.

Capture a named section:

agentshot "http://localhost:5173" "pricing.png" 
  --selector "section:has-text('Pricing')" --padding 24

Capture a specific repeated element:

agentshot "http://localhost:5173" "second-card.png" 
  --selector ".card" --nth 1 --padding 16

Wait for app-specific content before capturing:

agentshot "http://localhost:5173" "dashboard.png" 
  --wait-for "[data-ready='true']" --wait 500

The thing to check: can your agent capture the exact section it changed, inspect it, and continue without asking you to look?

5. Give your agent the prompt

The CLI package includes an agent instruction file. The short version is:

• After meaningful UI changes, run agentshot.
• Capture the changed route.
• Use --selector when only one section changed.
• Use --viewport 390x844 for mobile checks.
• Inspect the screenshot before claiming the UI is done.

This is the behavior we want to beta test: not just whether the screenshot saves, but whether the agent actually uses it to improve the work.

6. Send feedback

The fastest path is the CLI:

agentshot feedback "Selector capture worked, but full-page capture was slow on my app."

You can optionally tag feedback:

agentshot feedback "Chrome failed to launch on my CI image." --kind bug
agentshot feedback "Add a command that opens the latest screenshot." --kind idea

--kind only categorizes the message internally. Plain agentshot feedback "..." is enough.

What feedback is useful?

The useful reports are concrete:

• install command and OS
• URL type: public, staging, or localhost
• command you ran
• what you expected
• what happened
• whether a coding agent used the screenshot productively

The less polished the report, the better. We are trying to find the rough edges before the CLI becomes muscle memory.