Browser Automation with agent-browser
Quick start
bun x agent-browser open <url> # Navigate to page
bun x agent-browser snapshot -i # Get interactive elements with refs
bun x agent-browser click @e1 # Click element by ref
bun x agent-browser fill @e2 "text" # Fill input by ref
bun x agent-browser close # Close browser
Core workflow
- Navigate:
agent-browser open <url> - Snapshot:
agent-browser snapshot -i(returns elements with refs like@e1,@e2) - Interact using refs from the snapshot
- Re-snapshot after navigation or significant DOM changes
Commands
Navigation
bun x agent-browser open <url> # Navigate to URL
bun x agent-browser back # Go back
bun x agent-browser forward # Go forward
bun x agent-browser reload # Reload page
bun x agent-browser close # Close browser
Snapshot (page analysis)
bun x agent-browser snapshot # Full accessibility tree
bun x agent-browser snapshot -i # Interactive elements only (recommended)
bun x agent-browser snapshot -c # Compact output
bun x agent-browser snapshot -d 3 # Limit depth to 3
bun x agent-browser snapshot -s "#main" # Scope to CSS selector
Interactions (use @refs from snapshot)
bun x agent-browser click @e1 # Click
bun x agent-browser dblclick @e1 # Double-click
bun x agent-browser focus @e1 # Focus element
bun x agent-browser fill @e2 "text" # Clear and type
bun x agent-browser type @e2 "text" # Type without clearing
bun x agent-browser press Enter # Press key
bun x agent-browser press Control+a # Key combination
bun x agent-browser keydown Shift # Hold key down
bun x agent-browser keyup Shift # Release key
bun x agent-browser hover @e1 # Hover
bun x agent-browser check @e1 # Check checkbox
bun x agent-browser uncheck @e1 # Uncheck checkbox
bun x agent-browser select @e1 "value" # Select dropdown
bun x agent-browser scroll down 500 # Scroll page
bun x agent-browser scrollintoview @e1 # Scroll element into view
bun x agent-browser drag @e1 @e2 # Drag and drop
bun x agent-browser upload @e1 file.pdf # Upload files
Get information
bun x agent-browser get text @e1 # Get element text
bun x agent-browser get html @e1 # Get innerHTML
bun x agent-browser get value @e1 # Get input value
bun x agent-browser get attr @e1 href # Get attribute
bun x agent-browser get title # Get page title
bun x agent-browser get url # Get current URL
bun x agent-browser get count ".item" # Count matching elements
bun x agent-browser get box @e1 # Get bounding box
Check state
bun x agent-browser is visible @e1 # Check if visible
bun x agent-browser is enabled @e1 # Check if enabled
bun x agent-browser is checked @e1 # Check if checked
Screenshots & PDF
bun x agent-browser screenshot # Screenshot to stdout
bun x agent-browser screenshot path.png # Save to file
bun x agent-browser screenshot --full # Full page
bun x agent-browser pdf output.pdf # Save as PDF
Video recording
bun x agent-browser record start ./demo.webm # Start recording (uses current URL + state)
bun x agent-browser click @e1 # Perform actions
bun x agent-browser record stop # Stop and save video
bun x agent-browser record restart ./take2.webm # Stop current + start new recording
Recording creates a fresh context but preserves cookies/storage from your session. If no URL is provided, it automatically returns to your current page. For smooth demos, explore first, then start recording.
Wait
bun x agent-browser wait @e1 # Wait for element
bun x agent-browser wait 2000 # Wait milliseconds
bun x agent-browser wait --text "Success" # Wait for text
bun x agent-browser wait --url "**/dashboard" # Wait for URL pattern
bun x agent-browser wait --load networkidle # Wait for network idle
bun x agent-browser wait --fn "window.ready" # Wait for JS condition
Mouse control
bun x agent-browser mouse move 100 200 # Move mouse
bun x agent-browser mouse down left # Press button
bun x agent-browser mouse up left # Release button
bun x agent-browser mouse wheel 100 # Scroll wheel
Semantic locators (alternative to refs)
bun x agent-browser find role button click --name "Submit"
bun x agent-browser find text "Sign In" click
bun x agent-browser find label "Email" fill "user@test.com"
bun x agent-browser find first ".item" click
bun x agent-browser find nth 2 "a" text
Browser settings
bun x agent-browser set viewport 1920 1080 # Set viewport size
bun x agent-browser set device "iPhone 14" # Emulate device
bun x agent-browser set geo 37.7749 -122.4194 # Set geolocation
bun x agent-browser set offline on # Toggle offline mode
bun x agent-browser set headers '{"X-Key":"v"}' # Extra HTTP headers
bun x agent-browser set credentials user pass # HTTP basic auth
bun x agent-browser set media dark # Emulate color scheme
Cookies & Storage
bun x agent-browser cookies # Get all cookies
bun x agent-browser cookies set name value # Set cookie
bun x agent-browser cookies clear # Clear cookies
bun x agent-browser storage local # Get all localStorage
bun x agent-browser storage local key # Get specific key
bun x agent-browser storage local set k v # Set value
bun x agent-browser storage local clear # Clear all
Network
bun x agent-browser network route <url> # Intercept requests
bun x agent-browser network route <url> --abort # Block requests
bun x agent-browser network route <url> --body '{}' # Mock response
bun x agent-browser network unroute [url] # Remove routes
bun x agent-browser network requests # View tracked requests
bun x agent-browser network requests --filter api # Filter requests
Tabs & Windows
bun x agent-browser tab # List tabs
bun x agent-browser tab new [url] # New tab
bun x agent-browser tab 2 # Switch to tab
bun x agent-browser tab close # Close tab
bun x agent-browser window new # New window
Frames
bun x agent-browser frame "#iframe" # Switch to iframe
bun x agent-browser frame main # Back to main frame
Dialogs
bun x agent-browser dialog accept [text] # Accept dialog
bun x agent-browser dialog dismiss # Dismiss dialog
JavaScript
bun x agent-browser eval "document.title" # Run JavaScript
Example: Form submission
bun x agent-browser open https://example.com/form
bun x agent-browser snapshot -i
# Output shows: textbox "Email" [ref=e1], textbox "Password" [ref=e2], button "Submit" [ref=e3]
bun x agent-browser fill @e1 "user@example.com"
bun x agent-browser fill @e2 "password123"
bun x agent-browser click @e3
bun x agent-browser wait --load networkidle
bun x agent-browser snapshot -i # Check result
Example: Authentication with saved state
# Login once
bun x agent-browser open https://app.example.com/login
bun x agent-browser snapshot -i
bun x agent-browser fill @e1 "username"
bun x agent-browser fill @e2 "password"
bun x agent-browser click @e3
bun x agent-browser wait --url "**/dashboard"
bun x agent-browser state save auth.json
# Later sessions: load saved state
bun x agent-browser state load auth.json
bun x agent-browser open https://app.example.com/dashboard
Sessions (parallel browsers)
bun x agent-browser --session test1 open site-a.com
bun x agent-browser --session test2 open site-b.com
bun x agent-browser session list
JSON output (for parsing)
Add --json for machine-readable output:
bun x agent-browser snapshot -i --json
bun x agent-browser get text @e1 --json
Debugging
bun x agent-browser open example.com --headed # Show browser window
bun x agent-browser console # View console messages
bun x agent-browser errors # View page errors
bun x agent-browser record start ./debug.webm # Record from current page
bun x agent-browser record stop # Save recording
bun x agent-browser open example.com --headed # Show browser window
bun x agent-browser --cdp 9222 snapshot # Connect via CDP
bun x agent-browser console # View console messages
bun x agent-browser console --clear # Clear console
bun x agent-browser errors # View page errors
bun x agent-browser errors --clear # Clear errors
bun x agent-browser highlight @e1 # Highlight element
bun x agent-browser trace start # Start recording trace
bun x agent-browser trace stop trace.zip # Stop and save trace