25 min
Lesson 3 of 5
Web Automation with Computer Use
From the course Anthropic Computer Use Agent
Learn to automate web-based tasks using Claude's Computer Use capabilities, from simple navigation to complex multi-step workflows.
Basic Web Navigation
Opening and Controlling Browsers
def automate_web_task(task_description):
screenshot, size = capture_screen()
response = client.messages.create(
model="claude-3-5-sonnet-20241022",
max_tokens=1024,
tools=[{
"type": "computer_20241022",
"name": "computer",
"display_width_px": size[0],
"display_height_px": size[1],
}],
messages=[{
"role": "user",
"content": [
{"type": "text", "text": task_description},
{
"type": "image",
"source": {
"type": "base64",
"media_type": "image/png",
"data": screenshot
}
}
]
}]
)
# Execute the suggested actions
for content_block in response.content:
if content_block.type == "tool_use":
execute_computer_action(content_block.input)
return response
# Example usage
automate_web_task("Open a new browser tab and navigate to example.com")
Form Automation Examples
Contact Form Filling
def fill_contact_form(contact_data):
task = f"""
Fill out the contact form on this page with the following information:
Name: {contact_data['name']}
Email: {contact_data['email']}
Subject: {contact_data['subject']}
Message: {contact_data['message']}
Then click the submit button.
"""
return automate_web_task(task)
# Usage
contact_info = {
'name': 'John Doe',
'email': '[email protected]',
'subject': 'Product Inquiry',
'message': 'I am interested in learning more about your services.'
}
fill_contact_form(contact_info)
E-commerce Automation
def search_and_compare_products(search_query):
task = f"""
1. Search for '{search_query}' on this e-commerce site
2. Look at the first 5 results
3. For each product, note the name, price, and rating
4. Create a comparison of the options
"""
return automate_web_task(task)
# Example
search_and_compare_products("wireless headphones under $100")
Advanced Web Workflows
Multi-Tab Research
def research_topic(topic, sites):
task = f"""
Research '{topic}' by:
1. Opening separate tabs for each site: {', '.join(sites)}
2. Searching for the topic on each site
3. Collecting key information from the top 3 results on each
4. Summarizing the findings across all sources
"""
return automate_web_task(task)
# Usage
research_sites = [
"reddit.com",
"stackoverflow.com",
"medium.com"
]
research_topic("computer use automation", research_sites)
Data Extraction Pipeline
def extract_table_data(url, table_selector=None):
task = f"""
1. Navigate to {url}
2. Find the main data table on the page
3. Extract all rows and columns of data
4. Format the data as CSV or JSON
5. Copy the formatted data to clipboard
"""
if table_selector:
task += f"\nLook specifically for the table: {table_selector}"
return automate_web_task(task)
# Example
extract_table_data("https://example.com/data-page", "#main-table")
Error Handling and Reliability
Robust Task Execution
def execute_with_retry(task, max_attempts=3):
for attempt in range(max_attempts):
try:
result = automate_web_task(task)
# Check if task completed successfully
if "error" not in result.content[0].text.lower():
return result
except Exception as e:
print(f"Attempt {attempt + 1} failed: {e}")
if attempt < max_attempts - 1:
time.sleep(2) # Wait before retry
raise Exception("Task failed after all retry attempts")
Best Practices
Task Description Guidelines
- Be specific about the desired outcome
- Break complex tasks into clear steps
- Include context about what success looks like
- Specify data formats for extraction tasks
Performance Optimization
- Take screenshots only when necessary
- Use specific element descriptions to reduce processing time
- Batch related actions when possible
- Implement proper waiting mechanisms for page loads
Key Takeaways
- •Computer Use excels at web automation from simple navigation to complex workflows
- •Form filling and data extraction can be automated with natural language instructions
- •Multi-tab research and comparison tasks leverage Claude's ability to understand context
- •Robust error handling and retry logic improve reliability of automated tasks
0 of 5 lessons marked complete
Progress is kept in this browser only.