Browser Automation with agent-browser

Quick start

bash
1agent-browser open <url>        # Navigate to page
2agent-browser snapshot -i       # Get interactive elements with refs
3agent-browser click @e1         # Click element by ref
4agent-browser fill @e2 "text"   # Fill input by ref
5agent-browser close             # Close browser

Core workflow

1. Navigate: agent-browser open <url>
2. Snapshot: agent-browser snapshot -i (returns elements with refs like @e1, @e2)
3. Interact using refs from the snapshot
4. Re-snapshot after navigation or significant DOM changes

Commands

Navigation

bash
1agent-browser open <url>      # Navigate to URL
2agent-browser back            # Go back
3agent-browser forward         # Go forward
4agent-browser reload          # Reload page
5agent-browser close           # Close browser

Snapshot (page analysis)

bash
1agent-browser snapshot            # Full accessibility tree
2agent-browser snapshot -i         # Interactive elements only (recommended)
3agent-browser snapshot -c         # Compact output
4agent-browser snapshot -d 3       # Limit depth to 3
5agent-browser snapshot -s "#main" # Scope to CSS selector

Interactions (use @refs from snapshot)

bash
1agent-browser click @e1           # Click
2agent-browser dblclick @e1        # Double-click
3agent-browser focus @e1           # Focus element
4agent-browser fill @e2 "text"     # Clear and type
5agent-browser type @e2 "text"     # Type without clearing
6agent-browser press Enter         # Press key
7agent-browser press Control+a     # Key combination
8agent-browser keydown Shift       # Hold key down
9agent-browser keyup Shift         # Release key
10agent-browser hover @e1           # Hover
11agent-browser check @e1           # Check checkbox
12agent-browser uncheck @e1         # Uncheck checkbox
13agent-browser select @e1 "value"  # Select dropdown
14agent-browser scroll down 500     # Scroll page
15agent-browser scrollintoview @e1  # Scroll element into view
16agent-browser drag @e1 @e2        # Drag and drop
17agent-browser upload @e1 file.pdf # Upload files

Get information

bash
1agent-browser get text @e1        # Get element text
2agent-browser get html @e1        # Get innerHTML
3agent-browser get value @e1       # Get input value
4agent-browser get attr @e1 href   # Get attribute
5agent-browser get title           # Get page title
6agent-browser get url             # Get current URL
7agent-browser get count ".item"   # Count matching elements
8agent-browser get box @e1         # Get bounding box

Check state

bash
1agent-browser is visible @e1      # Check if visible
2agent-browser is enabled @e1      # Check if enabled
3agent-browser is checked @e1      # Check if checked

Data Extraction Best Practices

• Prefer innerText: Dynamic sites often obfuscate class names (e.g., css-1234). Use document.body.innerText to get all text and parse it with your LLM context.
• Use snapshot -i: This uses the accessibility tree which is more stable than HTML structure, especially for interactive elements.
• Avoid complex selectors: Selectors like div > div:nth-child(3) are brittle. If you must use JS, try robust queries like document.querySelectorAll('a[href*="showtime"]').
• Fallback: If a JS extraction returns [], immediately fallback to getting the whole page text.

Screenshots & PDF

bash
1agent-browser screenshot          # Screenshot to stdout
2agent-browser screenshot path.png # Save to file
3agent-browser screenshot --full   # Full page
4agent-browser pdf output.pdf      # Save as PDF

Video recording

bash
1agent-browser record start ./demo.webm    # Start recording (uses current URL + state)
2agent-browser click @e1                   # Perform actions
3agent-browser record stop                 # Stop and save video
4agent-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
1agent-browser wait @e1                     # Wait for element
2agent-browser wait 2000                    # Wait milliseconds
3agent-browser wait --text "Success"        # Wait for text
4agent-browser wait --url "**/dashboard"    # Wait for URL pattern
5agent-browser wait --load networkidle      # Wait for network idle
6agent-browser wait --fn "window.ready"     # Wait for JS condition

Mouse control

bash
1agent-browser mouse move 100 200      # Move mouse
2agent-browser mouse down left         # Press button
3agent-browser mouse up left           # Release button
4agent-browser mouse wheel 100         # Scroll wheel

Semantic locators (alternative to refs)

bash
1agent-browser find role button click --name "Submit"
2agent-browser find text "Sign In" click
3agent-browser find label "Email" fill "user@test.com"
4agent-browser find first ".item" click
5agent-browser find nth 2 "a" text

Browser settings

bash
1agent-browser set viewport 1920 1080      # Set viewport size
2agent-browser set device "iPhone 14"      # Emulate device
3agent-browser set geo 37.7749 -122.4194   # Set geolocation
4agent-browser set offline on              # Toggle offline mode
5agent-browser set headers '{"X-Key":"v"}' # Extra HTTP headers
6agent-browser set credentials user pass   # HTTP basic auth
7agent-browser set media dark              # Emulate color scheme

Cookies & Storage

bash
1agent-browser cookies                     # Get all cookies
2agent-browser cookies set name value      # Set cookie
3agent-browser cookies clear               # Clear cookies
4agent-browser storage local               # Get all localStorage
5agent-browser storage local key           # Get specific key
6agent-browser storage local set k v       # Set value
7agent-browser storage local clear         # Clear all

Network

bash
1agent-browser network route <url>              # Intercept requests
2agent-browser network route <url> --abort      # Block requests
3agent-browser network route <url> --body '{}'  # Mock response
4agent-browser network unroute [url]            # Remove routes
5agent-browser network requests                 # View tracked requests
6agent-browser network requests --filter api    # Filter requests

Tabs & Windows

bash
1agent-browser tab                 # List tabs
2agent-browser tab new [url]       # New tab
3agent-browser tab 2               # Switch to tab
4agent-browser tab close           # Close tab
5agent-browser window new          # New window

Frames

bash
1agent-browser frame "#iframe"     # Switch to iframe
2agent-browser frame main          # Back to main frame

Dialogs

bash
1agent-browser dialog accept [text]  # Accept dialog
2agent-browser dialog dismiss        # Dismiss dialog

JavaScript

bash
1agent-browser eval "document.title"   # Run JavaScript

Example: Form submission

bash
1agent-browser open https://example.com/form
2agent-browser snapshot -i
3# Output shows: textbox "Email" [ref=e1], textbox "Password" [ref=e2], button "Submit" [ref=e3]
4
5agent-browser fill @e1 "user@example.com"
6agent-browser fill @e2 "password123"
7agent-browser click @e3
8agent-browser wait --load networkidle
9agent-browser snapshot -i  # Check result

Example: Authentication with saved state

bash
1# Login once
2agent-browser open https://app.example.com/login
3agent-browser snapshot -i
4agent-browser fill @e1 "username"
5agent-browser fill @e2 "password"
6agent-browser click @e3
7agent-browser wait --url "**/dashboard"
8agent-browser state save auth.json
9
10# Later sessions: load saved state
11agent-browser state load auth.json
12agent-browser open https://app.example.com/dashboard

Sessions (parallel browsers)

bash
1agent-browser --session test1 open site-a.com
2agent-browser --session test2 open site-b.com
3agent-browser session list

JSON output (for parsing)

Add --json for machine-readable output:

bash
1agent-browser snapshot -i --json
2agent-browser get text @e1 --json

Debugging

bash
1agent-browser open example.com --headed              # Show browser window
2agent-browser console                                # View console messages
3agent-browser errors                                 # View page errors
4agent-browser record start ./debug.webm   # Record from current page
5agent-browser record stop                            # Save recording
6agent-browser open example.com --headed  # Show browser window
7agent-browser --cdp 9222 snapshot        # Connect via CDP
8agent-browser console                    # View console messages
9agent-browser console --clear            # Clear console
10agent-browser errors                     # View page errors
11agent-browser errors --clear             # Clear errors
12agent-browser highlight @e1              # Highlight element
13agent-browser trace start                # Start recording trace
14agent-browser trace stop trace.zip       # Stop and save trace