screenshot

screenshot

A Claude Code plugin that captures web page screenshots using Playwright. Designed for AI agent workflows where an agent needs to visually inspect rendered UI — take a screenshot, read the PNG, reason about layout and styling.

Supports viewport presets, authenticated sessions, dark mode, full-page capture, and HTML dumping.

Install

Clone and run directly:

git clone https://github.com/falense/claude-plugin-screenshot.git
cd claude-plugin-screenshot
uv run screenshot https://example.com

Chromium is installed automatically on first run.

As a Claude Code plugin

If your organization uses a Claude Code marketplace, you can add this plugin to your marketplace's marketplace.json:

{
  "name": "screenshot",
  "source": {
    "source": "github",
    "repo": "falense/claude-plugin-screenshot"
  },
  "description": "Capture screenshots of web pages using Playwright",
  "author": {
    "name": "Sondre"
  }
}

Then install with /plugin install screenshot@<your-marketplace>.

Usage

Basic screenshot

uv run screenshot https://example.com

Saves a PNG to .screenshots/ and prints the absolute path to stdout.

Viewport presets

uv run screenshot https://example.com --preset mobile
uv run screenshot https://example.com --preset tablet
uv run screenshot https://example.com --width 1920 --height 1080

Available presets:

Preset	Size
desktop (default)	1280×720
laptop	1440×900
tablet	768×1024
tablet-landscape	1024×768
mobile	375×667
mobile-large	414×896

Authenticated pages

Save a browser session by logging in interactively (opens a visible browser):

uv run screenshot login myapp https://myapp.com/login

Then reuse the session for headless captures:

uv run screenshot https://myapp.com/dashboard --session myapp

Sessions are stored in ~/.screenshot_sessions/ and persist across runs.

Additional options

uv run screenshot https://example.com --full-page          # capture entire scrollable page
uv run screenshot https://example.com --dark               # dark color scheme
uv run screenshot https://example.com --dump-html          # save rendered HTML alongside PNG
uv run screenshot https://example.com --click "#open-menu" # click element before capture
uv run screenshot https://example.com -o report.png        # explicit output path

Listing resources

uv run screenshot presets     # list viewport presets
uv run screenshot sessions    # list saved sessions

CLI reference

screenshot <url> [options]
screenshot login <name> <url>
screenshot sessions
screenshot presets

Options:
  -o, --output PATH         Output file path (default: .screenshots/<auto>.png)
  --width N                 Viewport width
  --height N                Viewport height
  --preset NAME             Named viewport preset
  --mobile                  Shortcut for --preset mobile
  --full-page               Capture full scrollable page
  --timeout MS              Navigation timeout (default: 30000)
  --session NAME            Use a saved session for authentication
  --dump-html               Save rendered HTML alongside the PNG
  --dark                    Capture in dark color scheme
  --click SELECTOR          Click an element before capture
  --click-delay SEC         Delay after click in seconds (default: 0.5)

Agent usage pattern

The tool is built for the agent loop: run the tool, read the image, reason about it.

# capture
uv run screenshot https://myapp.com/dashboard --session myapp

# the agent then reads the PNG to analyze layout, check for visual regressions, etc.

Output is deliberately minimal — just the file path on stdout, errors on stderr, non-zero exit on failure.

Development

uv run pytest              # run tests
uv run screenshot <url>    # test capture

Requirements

• Python 3.12+
• uv package manager