Vercel Browser Agent Browser

Capture accessibility tree of a page (93% smaller than screenshots).

# Basic snapshot
agent-browser snapshot https://example.com

# Output to file
agent-browser snapshot https://example.com > page.txt

# With element filter
agent-browser snapshot https://example.com --selector "main"

# JSON output
agent-browser snapshot https://example.com --format json

Output Format:

[1] link "Home"
[2] link "Products"
[3] button "Sign In"
[4] textbox "Email" value=""
[5] main
  [6] heading "Welcome"
  [7] paragraph "Get started..."

click

Click an element by text or selector.

# Click by visible text
agent-browser click "Sign In" --url https://example.com

# Click by role
agent-browser click "button:Submit" --url https://example.com

# Click by selector
agent-browser click "[data-testid='submit']" --url https://example.com

# Click nth element
agent-browser click "Add to Cart" --nth 0 --url https://shop.example.com

# Click with wait
agent-browser click "Continue" --wait 2000 --url https://example.com

fill

Fill form inputs with values.

# Fill by label
agent-browser fill "Email" "[email protected]" --url https://example.com

# Fill by placeholder
agent-browser fill "placeholder:Enter your email" "[email protected]" --url https://example.com

# Fill password
agent-browser fill "Password" "secret123" --url https://example.com

# Fill and submit
agent-browser fill "Search" "playwright testing" --submit --url https://google.com

# Clear before fill
agent-browser fill "Email" "[email protected]" --clear --url https://example.com

navigate

Navigate to URLs or within a page.

# Navigate to URL
agent-browser navigate https://example.com/dashboard

# Navigate with session
agent-browser navigate https://example.com --session myapp

# Go back
agent-browser back --url https://example.com

# Go forward
agent-browser forward --url https://example.com

# Reload
agent-browser reload --url https://example.com

wait

Wait for elements or conditions.

# Wait for element
agent-browser wait "Loading complete" --url https://example.com

# Wait for selector
agent-browser wait "[data-loaded='true']" --url https://example.com --timeout 10000

# Wait for navigation
agent-browser wait --navigation --url https://example.com

# Wait for network idle
agent-browser wait --network-idle --url https://example.com

type

Type text with keyboard simulation.

# Type text
agent-browser type "Hello World" --url https://example.com

# Type with delay (ms between keys)
agent-browser type "slow typing" --delay 100 --url https://example.com

# Press special keys
agent-browser press "Enter" --url https://example.com
agent-browser press "Tab" --url https://example.com
agent-browser press "Escape" --url https://example.com

# Key combinations
agent-browser press "Control+a" --url https://example.com
agent-browser press "Meta+c" --url https://example.com

select

Select dropdown options.

# Select by visible text
agent-browser select "Country" "United States" --url https://example.com

# Select by value
agent-browser select "Country" --value "US" --url https://example.com

# Select multiple (if supported)
agent-browser select "Toppings" "Cheese" "Pepperoni" --url https://example.com

scroll

Scroll the page or elements.

# Scroll down
agent-browser scroll down --url https://example.com

# Scroll up
agent-browser scroll up --url https://example.com

# Scroll to element
agent-browser scroll "Footer" --url https://example.com

# Scroll by pixels
agent-browser scroll --y 500 --url https://example.com

# Scroll to bottom
agent-browser scroll bottom --url https://example.com

extract

Extract data from pages.

# Extract text
agent-browser extract "h1" --url https://example.com

# Extract multiple
agent-browser extract ".product-name" --all --url https://shop.example.com

# Extract to JSON
agent-browser extract --schema '{"title": "h1", "price": ".price"}' --url https://example.com

# Extract table
agent-browser extract "table" --format csv --url https://example.com

Session Management

Create Session

# Create named session
agent-browser session create --name "myapp"

# Create with options
agent-browser session create --name "myapp" --persist --stealth

# List sessions
agent-browser session list

# Delete session
agent-browser session delete --name "myapp"

Use Session

# Use existing session
agent-browser --session "myapp" navigate https://example.com
agent-browser --session "myapp" click "Sign In"
agent-browser --session "myapp" fill "Email" "[email protected]"

# Session maintains cookies, localStorage, auth state

Cloud Mode (Browserbase)

Configuration

# Set environment variables
export BROWSERBASE_API_KEY="your-api-key"
export BROWSERBASE_PROJECT_ID="your-project-id"

# Or use CLI flags
agent-browser --cloud --api-key "..." snapshot https://example.com

Cloud Commands

# Use cloud browser
agent-browser --cloud snapshot https://example.com

# Cloud with stealth mode
agent-browser --cloud --stealth snapshot https://protected-site.com

# Cloud with proxy
agent-browser --cloud --proxy residential snapshot https://example.com

# Cloud with specific region
agent-browser --cloud --region us-east-1 snapshot https://example.com

Live Debugging

# Get live view URL
agent-browser --cloud session create --name "debug" --live

# Connect to view: https://browserbase.com/sessions/debug/live

Workflow Examples

Login Workflow

#!/bin/bash
# login.sh

URL="https://app.example.com"

# Navigate to login
agent-browser navigate "$URL/login"

# Fill credentials
agent-browser fill "Email" "$EMAIL"
agent-browser fill "Password" "$PASSWORD"

# Submit
agent-browser click "Sign In"

# Wait for redirect
agent-browser wait --navigation

# Verify login
agent-browser snapshot "$URL/dashboard" | grep "Welcome"

Form Submission

#!/bin/bash
# submit-form.sh

URL="https://example.com/contact"

agent-browser navigate "$URL"

agent-browser fill "Name" "John Doe"
agent-browser fill "Email" "[email protected]"
agent-browser fill "Message" "Hello, I have a question about your product."
agent-browser select "Subject" "Sales Inquiry"

agent-browser click "Submit"

agent-browser wait "Thank you"
echo "Form submitted successfully"

E-commerce Checkout

#!/bin/bash
# checkout.sh

BASE="https://shop.example.com"

# Add to cart
agent-browser navigate "$BASE/products/widget"
agent-browser click "Add to Cart"
agent-browser wait "Added to cart"

# Go to checkout
agent-browser navigate "$BASE/cart"
agent-browser click "Checkout"

# Fill shipping
agent-browser fill "Address" "123 Main St"
agent-browser fill "City" "San Francisco"
agent-browser fill "ZIP" "94102"
agent-browser select "State" "California"
agent-browser click "Continue"

# Fill payment (test mode)
agent-browser fill "Card Number" "4242424242424242"
agent-browser fill "Expiry" "12/30"
agent-browser fill "CVC" "123"

# Place order
agent-browser click "Place Order"
agent-browser wait "Order confirmed"

Data Extraction

#!/bin/bash
# scrape-products.sh

URL="https://shop.example.com/products"

agent-browser navigate "$URL"

# Extract all products
agent-browser extract --schema '{
  "products": [{
    "name": ".product-name",
    "price": ".product-price",
    "link": ".product-link@href"
  }]
}' --url "$URL" > products.json

Error Handling

Check Element Exists

# Check before click
if agent-browser snapshot "$URL" | grep -q "Sign In"; then
  agent-browser click "Sign In" --url "$URL"
else
  echo "Sign In button not found"
  exit 1
fi

Retry Logic

#!/bin/bash
# retry.sh

MAX_RETRIES=3
RETRY_DELAY=2

for i in $(seq 1 $MAX_RETRIES); do
  if agent-browser click "Submit" --url "$URL" 2>/dev/null; then
    echo "Success on attempt $i"
    break
  fi
  echo "Attempt $i failed, retrying in ${RETRY_DELAY}s..."
  sleep $RETRY_DELAY
done

Timeout Handling

# With timeout
timeout 30s agent-browser wait "Loading complete" --url "$URL" || echo "Timeout"

# CLI timeout flag
agent-browser click "Slow Button" --url "$URL" --timeout 15000

Options Reference

Deliverables

For: $ARGUMENTS

Provide:

1. Appropriate agent-browser command(s)
2. Session setup if multi-step
3. Error handling strategy
4. Expected output format
5. Cloud mode configuration if needed