Browser Automation with pnpx agent-browser

Quick start

bash

pnpx agent-browser open <url>        # Navigate to page
pnpx agent-browser snapshot -i       # Get interactive elements with refs
pnpx agent-browser click @e1         # Click element by ref
pnpx agent-browser fill @e2 "text"   # Fill input by ref
pnpx agent-browser close             # Close browser

Core workflow

•Navigate: pnpx agent-browser open <url>
•Snapshot: pnpx 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

bash

pnpx agent-browser open <url>      # Navigate to URL
pnpx agent-browser back            # Go back
pnpx agent-browser forward         # Go forward
pnpx agent-browser reload          # Reload page
pnpx agent-browser close           # Close browser

Snapshot (page analysis)

bash

pnpx agent-browser snapshot            # Full accessibility tree
pnpx agent-browser snapshot -i         # Interactive elements only (recommended)
pnpx agent-browser snapshot -c         # Compact output
pnpx agent-browser snapshot -d 3       # Limit depth to 3
pnpx agent-browser snapshot -s "#main" # Scope to CSS selector

Interactions (use @refs from snapshot)

bash

pnpx agent-browser click @e1           # Click
pnpx agent-browser dblclick @e1        # Double-click
pnpx agent-browser focus @e1           # Focus element
pnpx agent-browser fill @e2 "text"     # Clear and type
pnpx agent-browser type @e2 "text"     # Type without clearing
pnpx agent-browser press Enter         # Press key
pnpx agent-browser press Control+a     # Key combination
pnpx agent-browser keydown Shift       # Hold key down
pnpx agent-browser keyup Shift         # Release key
pnpx agent-browser hover @e1           # Hover
pnpx agent-browser check @e1           # Check checkbox
pnpx agent-browser uncheck @e1         # Uncheck checkbox
pnpx agent-browser select @e1 "value"  # Select dropdown
pnpx agent-browser scroll down 500     # Scroll page
pnpx agent-browser scrollintoview @e1  # Scroll element into view
pnpx agent-browser drag @e1 @e2        # Drag and drop
pnpx agent-browser upload @e1 file.pdf # Upload files

Get information

bash

pnpx agent-browser get text @e1        # Get element text
pnpx agent-browser get html @e1        # Get innerHTML
pnpx agent-browser get value @e1       # Get input value
pnpx agent-browser get attr @e1 href   # Get attribute
pnpx agent-browser get title           # Get page title
pnpx agent-browser get url             # Get current URL
pnpx agent-browser get count ".item"   # Count matching elements
pnpx agent-browser get box @e1         # Get bounding box

Check state

bash

pnpx agent-browser is visible @e1      # Check if visible
pnpx agent-browser is enabled @e1      # Check if enabled
pnpx agent-browser is checked @e1      # Check if checked

Screenshots & PDF

bash

pnpx agent-browser screenshot          # Screenshot to stdout
pnpx agent-browser screenshot path.png # Save to file
pnpx agent-browser screenshot --full   # Full page
pnpx agent-browser pdf output.pdf      # Save as PDF

Video recording

bash

pnpx agent-browser record start ./demo.webm    # Start recording (uses current URL + state)
pnpx agent-browser click @e1                   # Perform actions
pnpx agent-browser record stop                 # Stop and save video
pnpx 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

bash

pnpx agent-browser wait @e1                     # Wait for element
pnpx agent-browser wait 2000                    # Wait milliseconds
pnpx agent-browser wait --text "Success"        # Wait for text
pnpx agent-browser wait --url "**/dashboard"    # Wait for URL pattern
pnpx agent-browser wait --load networkidle      # Wait for network idle
pnpx agent-browser wait --fn "window.ready"     # Wait for JS condition

Mouse control

bash

pnpx agent-browser mouse move 100 200      # Move mouse
pnpx agent-browser mouse down left         # Press button
pnpx agent-browser mouse up left           # Release button
pnpx agent-browser mouse wheel 100         # Scroll wheel

Semantic locators (alternative to refs)

bash

pnpx agent-browser find role button click --name "Submit"
pnpx agent-browser find text "Sign In" click
pnpx agent-browser find label "Email" fill "user@test.com"
pnpx agent-browser find first ".item" click
pnpx agent-browser find nth 2 "a" text

Browser settings

bash

pnpx agent-browser set viewport 1920 1080      # Set viewport size
pnpx agent-browser set device "iPhone 14"      # Emulate device
pnpx agent-browser set geo 37.7749 -122.4194   # Set geolocation
pnpx agent-browser set offline on              # Toggle offline mode
pnpx agent-browser set headers '{"X-Key":"v"}' # Extra HTTP headers
pnpx agent-browser set credentials user pass   # HTTP basic auth
pnpx agent-browser set media dark              # Emulate color scheme

Cookies & Storage

bash

pnpx agent-browser cookies                     # Get all cookies
pnpx agent-browser cookies set name value      # Set cookie
pnpx agent-browser cookies clear               # Clear cookies
pnpx agent-browser storage local               # Get all localStorage
pnpx agent-browser storage local key           # Get specific key
pnpx agent-browser storage local set k v       # Set value
pnpx agent-browser storage local clear         # Clear all

Network

bash

pnpx agent-browser network route <url>              # Intercept requests
pnpx agent-browser network route <url> --abort      # Block requests
pnpx agent-browser network route <url> --body '{}'  # Mock response
pnpx agent-browser network unroute [url]            # Remove routes
pnpx agent-browser network requests                 # View tracked requests
pnpx agent-browser network requests --filter api    # Filter requests

Tabs & Windows

bash

pnpx agent-browser tab                 # List tabs
pnpx agent-browser tab new [url]       # New tab
pnpx agent-browser tab 2               # Switch to tab
pnpx agent-browser tab close           # Close tab
pnpx agent-browser window new          # New window

Frames

bash

pnpx agent-browser frame "#iframe"     # Switch to iframe
pnpx agent-browser frame main          # Back to main frame

Dialogs

bash

pnpx agent-browser dialog accept [text]  # Accept dialog
pnpx agent-browser dialog dismiss        # Dismiss dialog

JavaScript

bash

pnpx agent-browser eval "document.title"   # Run JavaScript

Example: Form submission

bash

pnpx agent-browser open https://example.com/form
pnpx agent-browser snapshot -i
# Output shows: textbox "Email" [ref=e1], textbox "Password" [ref=e2], button "Submit" [ref=e3]

pnpx agent-browser fill @e1 "user@example.com"
pnpx agent-browser fill @e2 "password123"
pnpx agent-browser click @e3
pnpx agent-browser wait --load networkidle
pnpx agent-browser snapshot -i  # Check result

Example: Authentication with saved state

bash

# Login once
pnpx agent-browser open https://app.example.com/login
pnpx agent-browser snapshot -i
pnpx agent-browser fill @e1 "username"
pnpx agent-browser fill @e2 "password"
pnpx agent-browser click @e3
pnpx agent-browser wait --url "**/dashboard"
pnpx agent-browser state save auth.json

# Later sessions: load saved state
pnpx agent-browser state load auth.json
pnpx agent-browser open https://app.example.com/dashboard

Sessions (parallel browsers)

bash

pnpx agent-browser --session test1 open site-a.com
pnpx agent-browser --session test2 open site-b.com
pnpx agent-browser session list

JSON output (for parsing)

Add --json for machine-readable output:

bash

pnpx agent-browser snapshot -i --json
pnpx agent-browser get text @e1 --json

Debugging

bash

pnpx agent-browser open example.com --headed              # Show browser window
pnpx agent-browser console                                # View console messages
pnpx agent-browser errors                                 # View page errors
pnpx agent-browser record start ./debug.webm   # Record from current page
pnpx agent-browser record stop                            # Save recording
pnpx agent-browser open example.com --headed  # Show browser window
pnpx agent-browser --cdp 9222 snapshot        # Connect via CDP
pnpx agent-browser console                    # View console messages
pnpx agent-browser console --clear            # Clear console
pnpx agent-browser errors                     # View page errors
pnpx agent-browser errors --clear             # Clear errors
pnpx agent-browser highlight @e1              # Highlight element
pnpx agent-browser trace start                # Start recording trace
pnpx agent-browser trace stop trace.zip       # Stop and save trace