The StagehandTool integrates the Stagehand framework with CrewAI, enabling agents to interact with websites and automate browser tasks using natural language instructions.
The context manager approach is recommended as it ensures proper cleanup of resources even if exceptions occur.
from crewai import Agent, Task, Crewfrom crewai_tools import StagehandToolfrom stagehand.schemas import AvailableModel# Initialize the tool with your API keys using a context managerwith StagehandTool( api_key="your-browserbase-api-key", project_id="your-browserbase-project-id", model_api_key="your-llm-api-key", # OpenAI or Anthropic API key model_name=AvailableModel.CLAUDE_3_7_SONNET_LATEST, # Optional: specify which model to use) as stagehand_tool: # Create an agent with the tool researcher = Agent( role="Web Researcher", goal="Find and summarize information from websites", backstory="I'm an expert at finding information online.", verbose=True, tools=[stagehand_tool], ) # Create a task that uses the tool research_task = Task( description="Go to https://www.example.com and tell me what you see on the homepage.", agent=researcher, ) # Run the crew crew = Crew( agents=[researcher], tasks=[research_task], verbose=True, ) result = crew.kickoff() print(result)
from crewai import Agent, Task, Crewfrom crewai_tools import StagehandToolfrom stagehand.schemas import AvailableModel# Initialize the tool with your API keysstagehand_tool = StagehandTool( api_key="your-browserbase-api-key", project_id="your-browserbase-project-id", model_api_key="your-llm-api-key", model_name=AvailableModel.CLAUDE_3_7_SONNET_LATEST,)try: # Create an agent with the tool researcher = Agent( role="Web Researcher", goal="Find and summarize information from websites", backstory="I'm an expert at finding information online.", verbose=True, tools=[stagehand_tool], ) # Create a task that uses the tool research_task = Task( description="Go to https://www.example.com and tell me what you see on the homepage.", agent=researcher, ) # Run the crew crew = Crew( agents=[researcher], tasks=[research_task], verbose=True, ) result = crew.kickoff() print(result)finally: # Explicitly clean up resources stagehand_tool.close()
The act command type (default) enables webpage interactions like clicking buttons, filling forms, and navigation.
# Perform an action (default behavior)result = stagehand_tool.run( instruction="Click the login button", url="https://example.com", command_type="act" # Default, so can be omitted)# Fill out a formresult = stagehand_tool.run( instruction="Fill the contact form with name 'John Doe', email 'john@example.com', and message 'Hello world'", url="https://example.com/contact")
The extract command type retrieves structured data from webpages.
# Extract all product informationresult = stagehand_tool.run( instruction="Extract all product names, prices, and descriptions", url="https://example.com/products", command_type="extract")# Extract specific information with a selectorresult = stagehand_tool.run( instruction="Extract the main article title and content", url="https://example.com/blog/article", command_type="extract", selector=".article-container" # Optional CSS selector)
The observe command type identifies and analyzes webpage elements.
# Find interactive elementsresult = stagehand_tool.run( instruction="Find all interactive elements in the navigation menu", url="https://example.com", command_type="observe")# Identify form fieldsresult = stagehand_tool.run( instruction="Identify all the input fields in the registration form", url="https://example.com/register", command_type="observe", selector="#registration-form")
Customize the StagehandTool behavior with these parameters:
stagehand_tool = StagehandTool( api_key="your-browserbase-api-key", project_id="your-browserbase-project-id", model_api_key="your-llm-api-key", model_name=AvailableModel.CLAUDE_3_7_SONNET_LATEST, dom_settle_timeout_ms=5000, # Wait longer for DOM to settle headless=True, # Run browser in headless mode self_heal=True, # Attempt to recover from errors wait_for_captcha_solves=True, # Wait for CAPTCHA solving verbose=1, # Control logging verbosity (0-3))