All posts

AgentScreenshots Features

A practical breakdown of the AgentScreenshots CLI features for local, preview, staging, and production visual checks.

positioning Miha Cacic May 21, 2026 4 min read

AgentScreenshots is a small product on purpose: install a CLI, point it at a page, save a screenshot locally, and let the coding agent inspect the rendered result.

This feature list is written for people deciding whether the workflow fits their frontend work.

Local-first screenshots

AgentScreenshots captures pages through a local-first CLI workflow.

That means it can capture local development URLs without a tunnel:

agentshot "http://localhost:5173" ".agents/screenshots/home.png"

It can also capture preview deployments, staging sites, and production pages reachable from the command environment:

agentshot "https://preview.example.com" ".agents/screenshots/preview-home.png"
agentshot "https://example.com" ".agents/screenshots/production-home.png"

The MVP does not upload your screenshots and does not run a hosted rendering service. The image file is saved at the output path you choose.

One CLI for agents

The npm package is named agentscreenshots. The command is agentshot.

npm install -g agentscreenshots
agentshot doctor
agentshot instructions

doctor checks whether the local runtime is ready. instructions gives coding agents the current operating guide, so teams do not have to paste a long prompt into every repository.

Full-page capture

For a normal page check, use a URL and an output file:

agentshot "http://localhost:5173" ".agents/screenshots/home-full.png"

For pages with lazy-loaded content or content that appears after scroll, add scroll and wait behavior:

agentshot "http://localhost:5173" ".agents/screenshots/home-scrolled.png" 
  --scroll 
  --wait 1000

This is useful when the agent needs to see the whole page before reporting back.

Section and selector capture

Most UI changes are local to one component or section. AgentScreenshots can capture just that area:

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

For repeated elements, use --nth:

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

Selector capture keeps the artifact focused. The agent can inspect the exact part of the interface it changed instead of searching through a long page screenshot.

Vertical slices

Sometimes a page is too tall, but a selector is not the right boundary. Capture a vertical range:

agentshot "http://localhost:5173" ".agents/screenshots/middle-band.png" 
  --from 1400 
  --to 2400

This works well for long marketing pages, documentation pages, and dashboards where a specific band needs review.

Mobile and desktop viewports

Responsive checks are a first-class workflow:

agentshot "http://localhost:5173" ".agents/screenshots/home-mobile.png" 
  --viewport 390x844 
  --scroll 
  --wait 1000

Desktop, tablet, and mobile screenshots can live next to each other in the same project folder:

agentshot "http://localhost:5173" ".agents/screenshots/home-desktop.png" 
  --viewport 1440x1000

agentshot "http://localhost:5173" ".agents/screenshots/home-mobile.png" 
  --viewport 390x844

This is one of the main reasons to give agents a screenshot tool. They can catch overflow, wrapping, and spacing issues before asking a human to review the work.

Waits and load states

Modern pages often need a moment before the right content is visible. AgentScreenshots supports explicit waits:

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

It can also wait for a browser load state:

agentshot "http://localhost:5173" ".agents/screenshots/home-loaded.png" 
  --wait-until load

The goal is not to hide broken loading behavior. The goal is to let the agent capture the intended state when the app has a clear readiness signal.

Click and hover states

Some visual states only appear after interaction.

Click before capture:

agentshot "http://localhost:5173/faq" ".agents/screenshots/faq-open.png" 
  --click "#security-toggle" 
  --selector "#faq"

Dismiss an optional banner without failing if it is absent:

agentshot "http://localhost:5173" ".agents/screenshots/home-clean.png" 
  --click-if-present "button:has-text('Reject all')" 
  --height 900

Capture a hover state:

agentshot "http://localhost:5173" ".agents/screenshots/nav-hover.png" 
  --hover ".nav-item" 
  --selector "nav" 
  --padding 20

Use --hover-if-present when the hover target may not exist on every page or breakpoint.

Device scale factor

When the agent needs a sharper image to inspect details, use a device scale factor:

agentshot "http://localhost:5173" ".agents/screenshots/hero@2x.png" 
  --selector ".hero" 
  --device-scale-factor 2

This can make small text, icons, and dense UI easier to review.

Local artifacts

Screenshots are ordinary files. A typical project convention is:

.agents/screenshots/

That makes captures easy to find, compare, and reference in final reports:

.agents/screenshots/pricing-desktop-after.png
.agents/screenshots/pricing-mobile-after.png

The artifact is the point. The agent should inspect the saved image before saying the UI is correct.

Usage and plans

AgentScreenshots counts successful screenshots as visual checks.

Failed captures do not count.

The anonymous local trial includes 50 one-time screenshots. The Free plan includes 100 visual checks per month. Paid plans are Solo, Pro, and Studio:

PlanMonthlyYearlyVisual checks
SoloEUR 5EUR 502,000/month
ProEUR 20EUR 20010,000/month
StudioEUR 100EUR 1,000100,000/month

The quota model is designed around visual checks, not seats.

What is intentionally not included

AgentScreenshots is not trying to cover every browser automation job.

The MVP does not include:

  • hosted screenshot rendering
  • screenshot uploads
  • a public screenshot API
  • scraping infrastructure
  • proxy or CAPTCHA tooling
  • browser-cloud sessions
  • visual regression baselines and diff approvals
  • MCP server behavior
  • general browser control

For multi-step browser operation, use a browser automation tool. For routine visual checks after frontend edits, AgentScreenshots is the smaller workflow.

Try it

Give your agent eyes in 30 seconds.

One CLI command. 100 visual checks free every month. No credit card.