selfevolveagent / evoagentx /tools /browser_tool.py
iLOVE2D's picture
Upload 2846 files
5374a2d verified
from selenium import webdriver
from selenium.webdriver.common.by import By
from selenium.webdriver.support.ui import WebDriverWait
from selenium.webdriver.support import expected_conditions as EC
from selenium.common.exceptions import TimeoutException, StaleElementReferenceException
from typing import Dict, Any, List, Optional, Tuple, Union
from pydantic import Field
from .tool import Tool,Toolkit
from ..core.module import BaseModule
from evoagentx.core.logging import logger
import html2text
import time
# Define selector map as a constant to avoid repetition
SELECTOR_MAP = {
"css": By.CSS_SELECTOR,
"xpath": By.XPATH,
"id": By.ID,
"class": By.CLASS_NAME,
"name": By.NAME,
"tag": By.TAG_NAME,
}
class BrowserBase(BaseModule):
"""
A tool for interacting with web browsers using Selenium.
Allows agents to navigate to URLs, interact with elements, extract information,
and more from web pages.
Key Features:
- Auto-initialization: Browser is automatically initialized when any method is first called
- Auto-cleanup: Browser is automatically closed when the instance is destroyed
- No manual initialization or cleanup required
"""
timeout: int = Field(default=10, description="Default timeout in seconds for browser operations")
browser_type: str = Field(default="chrome", description="Type of browser to use ('chrome', 'firefox', 'safari', 'edge')")
headless: bool = Field(default=False, description="Whether to run the browser in headless mode")
timeout: int = Field(default=10, description="Default timeout in seconds for browser operations")
def __init__(
self,
name: str = "Browser Tool",
browser_type: str = "chrome",
headless: bool = False,
timeout: int = 10,
**kwargs
):
"""
Initialize the browser tool with Selenium WebDriver.
Args:
name (str): Name of the tool
browser_type (str): Type of browser to use ('chrome', 'firefox', 'safari', 'edge')
headless (bool): Whether to run the browser in headless mode
timeout (int): Default timeout in seconds for browser operations
**kwargs: Additional keyword arguments for parent class initialization
"""
# Pass to parent class initialization
super().__init__(name=name, timeout=timeout, browser_type=browser_type, headless=headless, **kwargs)
self.driver = None
# Storage for element references from snapshots
self.element_references = {}
# Helper methods to reduce duplication
def _check_driver_initialized(self) -> Union[None, Dict[str, Any]]:
"""
Check if the browser driver is initialized. If not, initialize it automatically.
Returns:
Union[None, Dict[str, Any]]: None if driver is initialized, error response if initialization fails
"""
if not self.driver:
# Automatically initialize the browser
init_result = self.initialize_browser()
if init_result["status"] == "error":
return init_result
return None
def _get_selector_by_type(self, selector_type: str) -> Union[str, Dict[str, Any]]:
"""
Get the Selenium By selector for the given selector type.
Args:
selector_type (str): Type of selector ('css', 'xpath', 'id', 'class', 'name', 'tag')
Returns:
Union[str, Dict[str, Any]]: The By selector or error response
"""
by_type = SELECTOR_MAP.get(selector_type.lower())
if not by_type:
return {"status": "error", "message": f"Invalid selector type: {selector_type}"}
return by_type
def _wait_for_page_load(self, timeout: Optional[int] = None) -> bool:
"""
Wait for the page to load completely.
Args:
timeout (int, optional): Custom timeout for this operation
Returns:
bool: True if page loaded, False if timed out
"""
timeout = timeout or self.timeout
try:
WebDriverWait(self.driver, timeout).until(
lambda driver: driver.execute_script("return document.readyState") == "complete"
)
return True
except TimeoutException:
return False
def _parse_element_reference(self, ref: str) -> Tuple[Optional[str], Optional[str], Optional[str]]:
"""
Parse an element reference into selector type and selector.
Args:
ref (str): Element reference ID from the page snapshot
Returns:
Tuple[Optional[str], Optional[str], Optional[str]]:
(selector_type, selector, error_message) - error_message is None if successful
"""
if not self.element_references:
return None, None, "No page snapshot available. Use browser_snapshot or navigate_to_url first."
stored_ref = self.element_references.get(ref)
if not stored_ref:
return None, None, f"Element reference '{ref}' not found. Use browser_snapshot or navigate_to_url first."
# Parse the stored reference to get selector and type
if ":" in stored_ref:
ref_parts = stored_ref.split(":", 1)
if len(ref_parts) != 2:
return None, None, f"Invalid stored reference format: {stored_ref}"
selector_type, selector = ref_parts
return selector_type, selector, None
return None, None, f"Invalid stored reference format: {stored_ref}"
def _find_element_with_wait(self, by_type: str, selector: str,
timeout: Optional[int] = None,
wait_condition=EC.presence_of_element_located) -> Tuple[Optional[Any], Optional[str]]:
"""
Find an element on the page with wait condition.
Args:
by_type (str): Selenium By selector type
selector (str): The selector string
timeout (int, optional): Custom timeout for this operation
wait_condition: The EC condition to wait for
Returns:
Tuple[Optional[Any], Optional[str]]: (element, error_message) - error_message is None if successful
"""
timeout = timeout or self.timeout
try:
element = WebDriverWait(self.driver, timeout).until(
wait_condition((by_type, selector))
)
return element, None
except TimeoutException:
return None, f"Element not found or condition not met with selector: {selector}"
except Exception as e:
logger.error(f"Error finding element {selector}: {str(e)}")
return None, str(e)
def _handle_function_params(self, function_params: Optional[list],
function_name: str,
param_mapping: Dict[str, str]) -> Dict[str, Any]:
"""
Extract parameters from nested function_params format.
Args:
function_params (list, optional): Nested function parameters
function_name (str): The function name to look for
param_mapping (Dict[str, str]): Mapping of parameter names
Returns:
Dict[str, Any]: Extracted parameters
"""
result = {}
if not function_params:
return result
for param in function_params:
fn_name = param.get("function_name", "")
if fn_name == function_name or fn_name in param_mapping.get("alt_names", []):
args = param.get("function_args", {})
for param_name, result_name in param_mapping.items():
if param_name == "alt_names":
continue
if param_name in args:
result[result_name] = args[param_name]
break
return result
# Original methods with improved implementation using the helper methods
def initialize_browser(self, function_params: list = None) -> Dict[str, Any]:
"""
Start or restart a browser session. This method is called automatically when needed.
Note: This method is now called automatically by other browser methods when the browser
is not initialized. Manual initialization is no longer required.
This function supports multiple parameter styles:
1. Standard style: no parameters
2. Nested function_params style:
function_params=[{"function_name": "initialize_browser", "function_args": {}}]
Args:
function_params (list, optional): Nested function parameters
Returns:
Dict[str, Any]: Status information about the browser initialization
"""
try:
if self.driver:
# Close any existing session
try:
self.driver.quit()
except Exception as e:
logger.warning(f"Error closing existing browser session: {str(e)}")
options = None
if self.browser_type == "chrome":
from selenium.webdriver.chrome.options import Options
from selenium.webdriver.chrome.service import Service
from webdriver_manager.chrome import ChromeDriverManager
options = Options()
if self.headless:
options.add_argument("--headless")
options.add_argument("--no-sandbox")
options.add_argument("--disable-dev-shm-usage")
# Create service with Chrome executable path
service = Service(ChromeDriverManager().install())
self.driver = webdriver.Chrome(service=service, options=options)
elif self.browser_type == "firefox":
from selenium.webdriver.firefox.options import Options
options = Options()
if self.headless:
options.add_argument("--headless")
self.driver = webdriver.Firefox(options=options)
elif self.browser_type == "safari":
self.driver = webdriver.Safari()
elif self.browser_type == "edge":
from selenium.webdriver.edge.options import Options
options = Options()
if self.headless:
options.add_argument("--headless")
self.driver = webdriver.Edge(options=options)
else:
return {"status": "error", "message": f"Unsupported browser type: {self.browser_type}"}
self.driver.set_page_load_timeout(self.timeout)
return {"status": "success", "message": f"Browser {self.browser_type} initialized successfully"}
except Exception as e:
logger.error(f"Error initializing browser: {str(e)}")
return {"status": "error", "message": str(e)}
def navigate_to_url(self, url: str = None, timeout: int = None,
function_params: list = None) -> Dict[str, Any]:
"""
Navigate to a URL and capture a snapshot of the page. This provides element references used for interaction.
This function supports multiple parameter styles:
1. Standard style: url parameter
2. Nested function_params style:
function_params=[{"function_name": "navigate_to_url", "function_args": {"url": "..."}}]
Args:
url (str, optional): The complete URL (with https://) to navigate to
timeout (int, optional): Custom timeout in seconds (default: 10)
function_params (list, optional): Nested function parameters
Returns:
Dict[str, Any]: Information about the navigation result and page snapshot
"""
# Check if browser is initialized (will auto-initialize if needed)
driver_check = self._check_driver_initialized()
if driver_check:
return driver_check
# Handle nested function_params format
if function_params and not url:
params = self._handle_function_params(
function_params,
"navigate_to_url",
{"url": "url", "timeout": "timeout", "alt_names": ["browser_navigate"]}
)
url = params.get("url")
timeout = params.get("timeout", timeout)
if not url:
return {"status": "error", "message": "URL parameter is required"}
timeout = timeout or self.timeout
try:
self.driver.get(url)
# Wait for page to load
page_loaded = self._wait_for_page_load(timeout)
if not page_loaded:
logger.warning(f"Page load timeout for URL: {url}, but continuing with snapshot")
# Automatically take a snapshot of the page
snapshot_result = self.browser_snapshot()
if snapshot_result["status"] == "success":
return {
"status": "success",
"url": url,
"title": self.driver.title,
"current_url": self.driver.current_url,
"snapshot": {
"interactive_elements": snapshot_result.get("interactive_elements", [])
}
}
else:
# Return navigation success but note snapshot failure
return {
"status": "partial_success",
"url": url,
"title": self.driver.title,
"current_url": self.driver.current_url,
"snapshot_error": snapshot_result.get("message", "Unknown error capturing snapshot")
}
except TimeoutException:
return {"status": "timeout", "message": f"Timed out loading URL: {url}"}
except Exception as e:
logger.error(f"Error navigating to URL {url}: {str(e)}")
return {"status": "error", "message": str(e)}
def find_element(self, selector: str, selector_type: str = "css", timeout: int = None) -> Dict[str, Any]:
"""
Find an element on the current page and return information about it.
Args:
selector (str): The selector to find the element
selector_type (str): Type of selector ('css', 'xpath', 'id', 'class', 'name', 'tag')
timeout (int, optional): Custom timeout for this operation
Returns:
Dict[str, Any]: Information about the found element
"""
# Check if browser is initialized
driver_check = self._check_driver_initialized()
if driver_check:
return driver_check
timeout = timeout or self.timeout
# Get the selector type
by_type = self._get_selector_by_type(selector_type)
if isinstance(by_type, dict): # Error response
return by_type
try:
# Find the element
element, error = self._find_element_with_wait(
by_type, selector, timeout, EC.presence_of_element_located
)
if error:
return {"status": "not_found", "message": f"Element not found with {selector_type}: {selector}"}
# Extract element properties
element_properties = self._extract_element_properties(element, selector)
return {
"status": "success",
"element": element_properties
}
except Exception as e:
logger.error(f"Error finding element {selector}: {str(e)}")
return {"status": "error", "message": str(e)}
def _extract_element_properties(self, element, selector: str) -> Dict[str, Any]:
"""
Extract common properties from a WebElement.
Args:
element: The Selenium WebElement
selector (str): The selector used to find the element (for error messages)
Returns:
Dict[str, Any]: Element properties
"""
element_properties = {
"text": element.text,
"tag_name": element.tag_name,
"is_displayed": element.is_displayed(),
"is_enabled": element.is_enabled(),
}
# Get attributes safely
for attr in ["href", "id", "class"]:
try:
value = element.get_attribute(attr)
if value:
element_properties[attr] = value
except StaleElementReferenceException:
logger.warning(f"Element became stale when trying to get {attr} attribute for {selector}")
except Exception as e:
logger.warning(f"Could not get {attr} attribute for {selector}: {str(e)}")
return element_properties
def find_multiple_elements(self, selector: str, selector_type: str = "css", timeout: int = None) -> Dict[str, Any]:
"""
Find multiple elements on the current page and return information about them.
Args:
selector (str): The selector to find the elements
selector_type (str): Type of selector ('css', 'xpath', 'id', 'class', 'name', 'tag')
timeout (int, optional): Custom timeout for this operation
Returns:
Dict[str, Any]: Information about the found elements
"""
# Check if browser is initialized
driver_check = self._check_driver_initialized()
if driver_check:
return driver_check
timeout = timeout or self.timeout
# Get the selector type
by_type = self._get_selector_by_type(selector_type)
if isinstance(by_type, dict): # Error response
return by_type
try:
# First check if at least one element exists
element, error = self._find_element_with_wait(
by_type, selector, timeout, EC.presence_of_element_located
)
if error:
return {"status": "not_found", "message": f"No elements found with {selector_type}: {selector}"}
# Then get all matching elements
elements = self.driver.find_elements(by_type, selector)
# Extract element properties
elements_properties = []
for idx, element in enumerate(elements):
try:
element_properties = self._extract_element_properties(element, f"{selector}[{idx}]")
element_properties["index"] = idx
elements_properties.append(element_properties)
except StaleElementReferenceException:
logger.warning(f"Element {idx} became stale while extracting properties")
except Exception as e:
logger.warning(f"Error extracting properties for element {idx}: {str(e)}")
return {
"status": "success",
"count": len(elements_properties),
"elements": elements_properties
}
except Exception as e:
logger.error(f"Error finding elements {selector}: {str(e)}")
return {"status": "error", "message": str(e)}
def click_element(self, selector: str, selector_type: str = "css", timeout: int = None) -> Dict[str, Any]:
"""
Click on an element on the current page.
Args:
selector (str): The selector to find the element
selector_type (str): Type of selector ('css', 'xpath', 'id', 'class', 'name', 'tag')
timeout (int, optional): Custom timeout for this operation
Returns:
Dict[str, Any]: Result of the click operation
"""
# Check if browser is initialized
driver_check = self._check_driver_initialized()
if driver_check:
return driver_check
timeout = timeout or self.timeout
# Get the selector type
by_type = self._get_selector_by_type(selector_type)
if isinstance(by_type, dict): # Error response
return by_type
try:
# Find and click the element
element, error = self._find_element_with_wait(
by_type, selector, timeout, EC.element_to_be_clickable
)
if error:
return {"status": "not_found", "message": f"Element not clickable with {selector_type}: {selector}"}
element.click()
# Wait for page to load after click
page_loaded = self._wait_for_page_load(timeout)
if not page_loaded:
return {
"status": "partial_success",
"message": "Element clicked, but page load timed out",
"selector": selector,
"current_url": self.driver.current_url
}
return {
"status": "success",
"message": f"Clicked element with {selector_type}: {selector}",
"current_url": self.driver.current_url,
"title": self.driver.title
}
except Exception as e:
logger.error(f"Error clicking element {selector}: {str(e)}")
return {"status": "error", "message": str(e)}
def input_text(self, element: str = None, ref: str = None, text: str = None,
submit: bool = False, slowly: bool = True,
function_params: list = None) -> Dict[str, Any]:
"""
Type text into a form field, search box, or other input element using a reference ID from a snapshot.
This function only works with element references from a snapshot. Use browser_snapshot
or navigate_to_url first to capture the page elements.
This function supports multiple parameter styles:
1. Standard style: element (description), ref (element ID), text
2. Nested function_params style:
function_params=[{"function_name": "browser_type", "function_args": {...}}]
Args:
element (str, optional): Human-readable description of the element (e.g., 'Search field', 'Username input')
ref (str, optional): Element ID from the page snapshot (e.g., 'e0', 'e1', 'e2') - NOT a CSS selector
text (str, optional): Text to input into the element
submit (bool): Press Enter after typing to submit forms (default: false)
slowly (bool): Type one character at a time to trigger JS events (default: true)
function_params (list, optional): Nested function parameters
Returns:
Dict[str, Any]: Result of the text input operation
"""
# Check if browser is initialized
driver_check = self._check_driver_initialized()
if driver_check:
return driver_check
# Handle nested function_params format
if function_params:
params = self._handle_function_params(
function_params,
"input_text",
{"element": "element", "ref": "ref", "text": "text",
"submit": "submit", "slowly": "slowly", "alt_names": ["browser_type"]}
)
element = params.get("element", element)
ref = params.get("ref", ref)
text = params.get("text", text)
if "submit" in params:
submit = params["submit"]
if "slowly" in params:
slowly = params["slowly"]
if not ref or not text:
return {"status": "error", "message": "Both ref and text parameters are required"}
# Parse the reference
selector_type, selector, error = self._parse_element_reference(ref)
if error:
return {"status": "error", "message": error}
# Use a human-readable description or the ref ID if not provided
element_desc = element or ref
# Get the selector type
by_type = self._get_selector_by_type(selector_type)
if isinstance(by_type, dict): # Error response
return by_type
try:
# Find the element
web_element, error = self._find_element_with_wait(
by_type, selector, self.timeout, EC.element_to_be_clickable
)
if error:
return {"status": "not_found", "message": f"Element not found: {element_desc}"}
# Clear existing content
web_element.clear()
# Type text
if slowly:
# Type character by character
for char in text:
web_element.send_keys(char)
# Small delay between keypresses
time.sleep(0.05)
else:
# Type all at once
web_element.send_keys(text)
# Submit if requested
if submit:
from selenium.webdriver.common.keys import Keys
web_element.send_keys(Keys.ENTER)
# Wait for page to load after submission
page_loaded = self._wait_for_page_load(self.timeout)
if not page_loaded:
# Take a new snapshot after submitting
self.browser_snapshot()
return {
"status": "partial_success",
"message": "Text entered and submitted, but page load timed out",
"element": element_desc,
"text": text
}
# Take a new snapshot after submitting
snapshot_result = self.browser_snapshot()
if snapshot_result["status"] != "success":
logger.warning(f"Failed to capture snapshot after form submission: {snapshot_result.get('message')}")
return {
"status": "success",
"message": f"Successfully input text into {element_desc}" +
(" and submitted" if submit else ""),
"element": element_desc,
"text": text
}
except TimeoutException:
return {"status": "not_found", "message": f"Element not found: {element_desc}"}
except Exception as e:
logger.error(f"Error inputting text to element {element_desc}: {str(e)}")
return {"status": "error", "message": str(e)}
def get_page_content(self) -> Dict[str, Any]:
"""
Get the current page title, URL and body content.
Returns:
Dict[str, Any]: Information about the current page
"""
# Check if browser is initialized (will auto-initialize if needed)
driver_check = self._check_driver_initialized()
if driver_check:
return driver_check
try:
# Get title and URL
title = self.driver.title
current_url = self.driver.current_url
# Extract only the body content using JavaScript
body_content = self.driver.execute_script("""
var body = document.body;
return body ? body.outerHTML : "";
""")
# Get a summary of key elements for easier navigation
element_summary = self.driver.execute_script("""
// Get common interactive elements
var summary = {
links: [],
buttons: [],
inputs: [],
forms: []
};
// Get links
var links = document.querySelectorAll('a');
for (var i = 0; i < Math.min(links.length, 20); i++) {
var link = links[i];
summary.links.push({
text: link.textContent.trim().substring(0, 50),
href: link.getAttribute('href'),
id: link.id,
class: link.className
});
}
// Get buttons
var buttons = document.querySelectorAll('button, input[type="button"], input[type="submit"]');
for (var i = 0; i < Math.min(buttons.length, 20); i++) {
var button = buttons[i];
summary.buttons.push({
text: button.textContent ? button.textContent.trim().substring(0, 50) : button.value,
id: button.id,
class: button.className,
type: button.type
});
}
// Get inputs
var inputs = document.querySelectorAll('input:not([type="button"]):not([type="submit"]), textarea, select');
for (var i = 0; i < Math.min(inputs.length, 20); i++) {
var input = inputs[i];
summary.inputs.push({
type: input.type,
name: input.name,
id: input.id,
placeholder: input.placeholder
});
}
// Get forms
var forms = document.querySelectorAll('form');
for (var i = 0; i < Math.min(forms.length, 10); i++) {
var form = forms[i];
summary.forms.push({
id: form.id,
action: form.action,
method: form.method
});
}
return summary;
""")
return {
"status": "success",
"title": title,
"url": current_url,
"body_content": body_content,
"element_summary": element_summary
}
except Exception as e:
logger.error(f"Error getting page content: {str(e)}")
return {"status": "error", "message": str(e)}
def switch_to_frame(self, frame_reference: str, reference_type: str = "index") -> Dict[str, Any]:
"""
Switch to a frame on the page.
Args:
frame_reference (str): Reference to the frame (index, name, or ID)
reference_type (str): Type of reference ('index', 'name', 'id', 'element')
Returns:
Dict[str, Any]: Result of the frame switch operation
"""
# Check if browser is initialized (will auto-initialize if needed)
driver_check = self._check_driver_initialized()
if driver_check:
return driver_check
try:
if reference_type == "index":
try:
index = int(frame_reference)
self.driver.switch_to.frame(index)
except ValueError:
return {"status": "error", "message": f"Invalid frame index: {frame_reference}"}
elif reference_type == "name" or reference_type == "id":
self.driver.switch_to.frame(frame_reference)
elif reference_type == "element":
# First find the element
selector_parts = frame_reference.split(":", 1)
if len(selector_parts) != 2:
return {"status": "error", "message": "Element reference must be in format 'selector_type:selector'"}
selector_type, selector = selector_parts
element_result = self.find_element(selector, selector_type)
if element_result["status"] != "success":
return {"status": "error", "message": f"Could not find frame element: {element_result['message']}"}
# Get the actual WebElement (not just the properties)
selector_map = {
"css": By.CSS_SELECTOR,
"xpath": By.XPATH,
"id": By.ID,
"class": By.CLASS_NAME,
"name": By.NAME,
"tag": By.TAG_NAME,
}
by_type = selector_map.get(selector_type.lower())
element = self.driver.find_element(by_type, selector)
self.driver.switch_to.frame(element)
else:
return {"status": "error", "message": f"Invalid reference type: {reference_type}"}
return {
"status": "success",
"message": f"Switched to frame using {reference_type}: {frame_reference}"
}
except Exception as e:
logger.error(f"Error switching to frame {frame_reference}: {str(e)}")
return {"status": "error", "message": str(e)}
def switch_to_window(self, window_reference: str, reference_type: str = "index") -> Dict[str, Any]:
"""
Switch to a window or tab.
Args:
window_reference (str): Reference to the window (index, handle, or title)
reference_type (str): Type of reference ('index', 'handle', 'title')
Returns:
Dict[str, Any]: Result of the window switch operation
"""
# Check if browser is initialized (will auto-initialize if needed)
driver_check = self._check_driver_initialized()
if driver_check:
return driver_check
try:
window_handles = self.driver.window_handles
if not window_handles:
return {"status": "error", "message": "No window handles available"}
if reference_type == "index":
try:
index = int(window_reference)
if index < 0 or index >= len(window_handles):
return {"status": "error", "message": f"Window index out of range: {index}"}
self.driver.switch_to.window(window_handles[index])
except ValueError:
return {"status": "error", "message": f"Invalid window index: {window_reference}"}
elif reference_type == "handle":
if window_reference not in window_handles:
return {"status": "error", "message": f"Window handle not found: {window_reference}"}
self.driver.switch_to.window(window_reference)
elif reference_type == "title":
current_handle = self.driver.current_window_handle
window_found = False
for handle in window_handles:
try:
self.driver.switch_to.window(handle)
if self.driver.title == window_reference:
window_found = True
break
except Exception:
pass
if not window_found:
# Switch back to the original window
self.driver.switch_to.window(current_handle)
return {"status": "error", "message": f"No window with title '{window_reference}' found"}
else:
return {"status": "error", "message": f"Invalid reference type: {reference_type}"}
return {
"status": "success",
"message": f"Switched to window using {reference_type}: {window_reference}",
"title": self.driver.title,
"url": self.driver.current_url
}
except Exception as e:
logger.error(f"Error switching to window {window_reference}: {str(e)}")
return {"status": "error", "message": str(e)}
def select_dropdown_option(self, select_selector: str,
option_value: str,
select_by: str = "value",
selector_type: str = "css") -> Dict[str, Any]:
"""
Select an option from a dropdown
select_by can be 'value', 'text', or 'index'
Args:
select_selector (str): The selector to find the dropdown element
option_value (str): The value to select (depends on select_by)
select_by (str): Method to select by ('value', 'text', 'index')
selector_type (str): Type of selector for the dropdown
Returns:
Dict[str, Any]: Result of the selection operation
"""
# Check if browser is initialized
driver_check = self._check_driver_initialized()
if driver_check:
return driver_check
try:
from selenium.webdriver.support.ui import Select
# Get the selector type
by_type = self._get_selector_by_type(selector_type)
if isinstance(by_type, dict): # Error response
return by_type
# Find the dropdown element
element, error = self._find_element_with_wait(
by_type, select_selector, self.timeout, EC.presence_of_element_located
)
if error:
return {"status": "not_found", "message": f"Dropdown element not found with {selector_type}: {select_selector}"}
# Create select object
select = Select(element)
# Select based on method
if select_by.lower() == "value":
select.select_by_value(option_value)
elif select_by.lower() == "text":
select.select_by_visible_text(option_value)
elif select_by.lower() == "index":
try:
select.select_by_index(int(option_value))
except ValueError:
return {"status": "error", "message": f"Invalid index value: {option_value}. Must be an integer."}
else:
return {"status": "error", "message": f"Invalid select_by option: {select_by}"}
return {"status": "success", "message": f"Selected option with {select_by}: {option_value}"}
except Exception as e:
logger.error(f"Error selecting dropdown option: {str(e)}")
return {"status": "error", "message": str(e)}
def close_browser(self) -> Dict[str, Any]:
"""
Close the browser and end the session. Call this when you're done to free resources.
Returns:
Dict[str, Any]: Status of the browser closure
"""
if not self.driver:
return {"status": "success", "message": "Browser already closed"}
try:
self.driver.quit()
self.driver = None
return {"status": "success", "message": "Browser closed successfully"}
except Exception as e:
logger.error(f"Error closing browser: {str(e)}")
return {"status": "error", "message": str(e)}
def browser_click(self, element: str = None, ref: str = None,
function_params: list = None) -> Dict[str, Any]:
"""
Click on a button, link, or other clickable element using a reference ID from a snapshot.
This function only works with element references from a snapshot. You MUST call browser_snapshot
or navigate_to_url first to capture the page elements.
Common usage pattern:
1. First get a snapshot: browser_snapshot() or navigate_to_url()
2. Find the element reference (e.g. 'e0', 'e1') from the snapshot's interactive_elements
3. Use that reference to click: browser_click(element='Login button', ref='e0')
This function supports multiple parameter styles:
1. Standard style: element (description), ref (element ID)
2. Nested function_params style:
function_params=[{"function_name": "browser_click", "function_args": {...}}]
Args:
element (str, optional): Human-readable description of what you're clicking (e.g., 'Login button', 'Next page link')
ref (str, optional): Element ID from the page snapshot (e.g., 'e0', 'e1', 'e2') - NOT a CSS selector
function_params (list, optional): Nested function parameters
Returns:
Dict[str, Any]: Result of the click operation with detailed feedback
"""
# Check if browser is initialized
driver_check = self._check_driver_initialized()
if driver_check:
return driver_check
# Handle nested function_params format
if function_params and not ref:
params = self._handle_function_params(
function_params,
"browser_click",
{"element": "element", "ref": "ref"}
)
element = params.get("element", element)
ref = params.get("ref", ref)
# Validate required parameters
if not ref:
return {
"status": "error",
"message": "Element reference (ref) parameter is required. You must first call browser_snapshot() or navigate_to_url() to get element references.",
"required_steps": [
"1. Call browser_snapshot() or navigate_to_url() to get page elements",
"2. Find the element reference (e.g. 'e0') in the response's interactive_elements",
"3. Use that reference to click: browser_click(element='Button name', ref='e0')"
]
}
# Check if we have any stored references
if not self.element_references:
return {
"status": "error",
"message": "No element references found. You must first capture a page snapshot.",
"required_steps": [
"1. Call browser_snapshot() or navigate_to_url() to capture the page state",
"2. Use the element references returned in the snapshot"
]
}
# Parse the reference
selector_type, selector, error = self._parse_element_reference(ref)
if error:
return {
"status": "error",
"message": error,
"help": "Make sure you're using a valid element reference from a recent snapshot"
}
# Use a human-readable description or the ref ID if not provided
element_desc = element or ref
# Get the selector type
by_type = self._get_selector_by_type(selector_type)
if isinstance(by_type, dict): # Error response
return by_type
try:
# First check if element exists at all
try:
element_exists = self.driver.find_element(by_type, selector)
except Exception:
return {
"status": "not_found",
"message": f"Element not found: {element_desc}",
"suggestion": "The page may have changed. Try getting a new snapshot with browser_snapshot()"
}
# Then check if it's clickable
web_element, error = self._find_element_with_wait(
by_type, selector, self.timeout, EC.element_to_be_clickable
)
if error:
# Element exists but isn't clickable - get more details
try:
is_visible = element_exists.is_displayed()
is_enabled = element_exists.is_enabled()
element_tag = element_exists.tag_name
element_classes = element_exists.get_attribute("class")
return {
"status": "not_clickable",
"message": f"Element found but not clickable: {element_desc}",
"element_state": {
"visible": is_visible,
"enabled": is_enabled,
"tag": element_tag,
"classes": element_classes
},
"suggestion": "The element might be disabled, hidden, or covered by another element"
}
except Exception:
return {
"status": "not_clickable",
"message": f"Element found but not clickable: {element_desc}",
"suggestion": "The element might be disabled, hidden, or covered by another element"
}
# Try to click the element
web_element.click()
# Wait for page to load after click
page_loaded = self._wait_for_page_load(self.timeout)
if not page_loaded:
# Take a snapshot anyway even if page load times out
snapshot_result = self.browser_snapshot()
return {
"status": "partial_success",
"message": "Element clicked, but page load timed out",
"element": element_desc,
"current_url": self.driver.current_url,
"snapshot": snapshot_result if snapshot_result["status"] == "success" else None,
"suggestion": "The page might still be loading. You may want to wait and take another snapshot."
}
# Take a new snapshot after the click
snapshot_result = self.browser_snapshot()
if snapshot_result["status"] == "success":
return {
"status": "success",
"message": f"Successfully clicked on {element_desc}",
"element": element_desc,
"current_url": self.driver.current_url,
"title": self.driver.title,
"snapshot": {
"interactive_elements": snapshot_result.get("interactive_elements", [])
}
}
else:
# Return click success but note snapshot failure
return {
"status": "success",
"message": f"Successfully clicked on {element_desc} but snapshot failed",
"element": element_desc,
"current_url": self.driver.current_url,
"title": self.driver.title,
"snapshot_error": snapshot_result.get("message", "Unknown error capturing snapshot"),
"suggestion": "You may want to take another snapshot with browser_snapshot()"
}
except TimeoutException:
return {
"status": "timeout",
"message": f"Timed out waiting for element to be clickable: {element_desc}",
"suggestion": "The element might be taking too long to load or become clickable"
}
except Exception as e:
logger.error(f"Error clicking element: {str(e)}")
return {
"status": "error",
"message": str(e),
"element": element_desc,
"suggestion": "Try getting a new snapshot of the page with browser_snapshot()"
}
def _classify_element_interactivity(self, element_data: Dict[str, Any]) -> Dict[str, Any]:
"""
Classify an element's interactivity based on its properties.
This method contains all rules for determining if an element is interactive or editable.
Args:
element_data (Dict[str, Any]): Element data including properties, attributes, etc.
Returns:
Dict[str, Any]: Element data with interactivity classifications added
"""
# Start with basic properties
element_data["interactable"] = False
element_data["editable"] = False
# Get commonly used properties
tag_name = element_data.get("properties", {}).get("tag", "").upper()
role = element_data.get("attributes", {}).get("role", "").lower()
# Check if element is disabled
is_disabled = (
element_data.get("attributes", {}).get("disabled") is not None or
element_data.get("attributes", {}).get("aria-disabled") == "true" or
element_data.get("attributes", {}).get("aria-hidden") == "true"
)
# Check visibility
is_visible = element_data.get("visible", True)
if not is_disabled and is_visible:
# Interactive Tags Check
interactive_tags = {
'A', 'BUTTON', 'INPUT', 'SELECT', 'TEXTAREA',
'DETAILS', 'AUDIO', 'VIDEO', 'IFRAME', 'EMBED',
'OBJECT', 'SUMMARY', 'MENU'
}
# Interactive Roles Check
interactive_roles = {
'button', 'link', 'checkbox', 'menuitem',
'menuitemcheckbox', 'menuitemradio', 'option',
'radio', 'searchbox', 'slider', 'spinbutton',
'switch', 'tab', 'textbox', 'combobox',
'listbox', 'menu', 'menubar', 'radiogroup',
'tablist', 'toolbar', 'tree', 'treegrid'
}
# Check for interactive attributes
has_interactive_attrs = any([
element_data.get("attributes", {}).get(attr) is not None
for attr in ['onclick', 'onkeydown', 'onkeyup', 'onmousedown',
'onmouseup', 'tabindex']
])
# Determine interactability
element_data["interactable"] = (
tag_name in interactive_tags or
role in interactive_roles or
has_interactive_attrs
)
# Determine editability
editable_input_types = {'text', 'search', 'email', 'number', 'tel',
'url', 'password'}
editable_roles = {'textbox', 'searchbox', 'spinbutton'}
element_data["editable"] = (
# Standard input fields
(tag_name == 'INPUT' and
element_data.get("attributes", {}).get("type", "text").lower() in editable_input_types) or
tag_name == 'TEXTAREA' or
# Rich text editing
element_data.get("attributes", {}).get("contenteditable") == "true" or
# Explicit input roles
role in editable_roles
)
return element_data
def _process_accessibility_tree(self, accessibility_tree):
"""
Process the accessibility tree to extract all elements and store their references.
This method processes all elements in the page structure, assigns unique IDs,
and stores their selectors for later interaction.
Args:
accessibility_tree (dict): The accessibility tree from JavaScript
Returns:
list: A list of all elements with their IDs and properties
"""
all_elements = []
# Function to extract all elements and store references
def extract_elements(node, path="", index=0):
if not node:
return index
current_path = path + "/" + (node.get("name") or node.get("role") or "element")
# Generate a unique element ID for all elements
element_id = f"e{index}"
element_info = {
"id": element_id,
"description": current_path.strip("/"),
"purpose": node.get("semantic_info", {}).get("purpose", ""),
"label": node.get("semantic_info", {}).get("label", ""),
"category": node.get("semantic_info", {}).get("category", ""),
"isPrimary": node.get("semantic_info", {}).get("isPrimary", False),
"visible": node.get("visible", True),
"properties": node.get("properties", {}),
"attributes": node.get("attributes", {})
}
# Store the first selector as the actual reference for this element
if "all_refs" in node:
self.element_references[element_id] = node["all_refs"][0]
# Classify element interactivity
element_info = self._classify_element_interactivity(element_info)
all_elements.append(element_info)
index += 1
# Process children
for child in node.get("children", []):
index = extract_elements(child, current_path, index)
return index
# Extract all elements
extract_elements(accessibility_tree)
return all_elements
def browser_snapshot(self, function_params: list = None) -> Dict[str, Any]:
"""
Capture a fresh snapshot of the current page with all interactive elements.
Use after page state changes not caused by navigation or clicking.
This function supports multiple parameter styles:
1. Standard style: no parameters
2. Nested function_params style:
function_params=[{"function_name": "browser_snapshot", "function_args": {}}]
Args:
function_params (list, optional): Nested function parameters
Returns:
Dict[str, Any]: The accessibility snapshot of the page with interactive elements
"""
# Check if browser is initialized
driver_check = self._check_driver_initialized()
if driver_check:
return driver_check
try:
# Get basic page info
title = self.driver.title
current_url = self.driver.current_url
# Get accessibility tree using JavaScript
accessibility_tree = self.driver.execute_script("""
function getAccessibilityTree(node, depth = 0, maxDepth = 10) {
if (!node || depth > maxDepth) return null;
let result = {
role: node.role || node.tagName,
name: node.name || '',
type: node.type || '',
value: node.value || '',
description: node.description || '',
properties: {},
visible: isElementVisible(node)
};
// Helper function for element visibility
function isElementVisible(element) {
if (!element.getBoundingClientRect) return true;
const style = window.getComputedStyle(element);
const rect = element.getBoundingClientRect();
// Check basic visibility
const isVisible = style.display !== 'none' &&
style.visibility !== 'hidden' &&
style.opacity !== '0' &&
rect.width > 0 &&
rect.height > 0;
// Check if element is in viewport
const isInViewport = rect.top >= 0 &&
rect.left >= 0 &&
rect.bottom <= window.innerHeight &&
rect.right <= window.innerWidth;
return isVisible && isInViewport;
}
// Add text content
if (node.textContent) {
result.text_content = node.textContent.trim();
}
// Add identifier properties for references
if (node.id) result.properties.id = node.id;
if (node.className) result.properties.class = node.className;
if (node.tagName) result.properties.tag = node.tagName.toLowerCase();
// Add attributes
if (node.attributes) {
result.attributes = {};
for (let attr of node.attributes) {
result.attributes[attr.name] = attr.value;
}
}
// Add custom ref property that combines selector types
let refs = [];
// Store all possible selectors, but don't use them as primary ref
if (node.id) refs.push(`id:${node.id}`);
if (node.className && typeof node.className === 'string')
refs.push(`class:${node.className}`);
if (node.tagName) refs.push(`tag:${node.tagName.toLowerCase()}`);
// For inputs, add name attribute
if (node.getAttribute && node.getAttribute('name')) {
result.properties.name = node.getAttribute('name');
refs.push(`name:${node.getAttribute('name')}`);
}
// Create XPath and CSS selectors
try {
// CSS selector
let cssPath = getCssPath(node);
if (cssPath) refs.push(`css:${cssPath}`);
// XPath
let xpath = getXPath(node);
if (xpath) refs.push(`xpath:${xpath}`);
} catch (e) {}
// Store all refs but don't set primary ref here
if (refs.length > 0) {
result.all_refs = refs;
}
// Add semantic information about the element
result.semantic_info = {
// What the element represents
purpose: (function() {
if (node.tagName === 'INPUT') {
if (node.type === 'submit') return 'submit button';
if (node.type === 'search') return 'search box';
if (node.type === 'text') return 'text input';
return `${node.type || 'text'} input`;
}
if (node.tagName === 'BUTTON') return 'button';
if (node.tagName === 'A') return 'link';
if (node.tagName === 'SELECT') return 'dropdown';
if (node.tagName === 'TEXTAREA') return 'text area';
if (node.getAttribute('role')) return node.getAttribute('role');
return 'interactive element';
})(),
// The visible or accessible text
label: (function() {
return node.getAttribute('aria-label') ||
node.getAttribute('title') ||
node.getAttribute('placeholder') ||
node.getAttribute('alt') ||
(node.tagName === 'INPUT' ? node.value : node.textContent.trim());
})(),
// Is this a primary action?
isPrimary: !!(
node.classList.contains('primary') ||
node.getAttribute('aria-label')?.toLowerCase().includes('search') ||
node.getAttribute('title')?.toLowerCase().includes('search') ||
node.type === 'search' ||
node.getAttribute('role') === 'main' ||
node.id?.toLowerCase().includes('main') ||
node.classList.contains('main')
),
// Basic category
category: (function() {
if (node.type === 'search' ||
node.getAttribute('role') === 'searchbox') return 'search';
if (node.type === 'submit' ||
node.tagName === 'BUTTON' ||
node.getAttribute('role') === 'button') return 'action';
if (node.tagName === 'A' ||
node.getAttribute('role') === 'link') return 'navigation';
if (node.tagName === 'INPUT' ||
node.tagName === 'TEXTAREA' ||
node.getAttribute('role') === 'textbox') return 'input';
if (node.tagName === 'SELECT' ||
['listbox', 'combobox'].includes(node.getAttribute('role'))) return 'selection';
return 'interactive';
})()
};
// Process children
result.children = [];
if (node.children) {
for (let i = 0; i < node.children.length; i++) {
const childTree = getAccessibilityTree(node.children[i], depth + 1, maxDepth);
if (childTree) {
result.children.push(childTree);
}
}
}
return result;
}
return getAccessibilityTree(document.body);
""")
# Process the accessibility tree and extract all elements
all_elements = self._process_accessibility_tree(accessibility_tree)
page_content = html2text.html2text(self.driver.page_source)
return {
"status": "success",
"title": title,
"url": current_url,
"accessibility_tree": accessibility_tree,
"page_content": page_content,
"interactive_elements": [e for e in all_elements if e.get("interactable") or e.get("editable")]
}
except Exception as e:
logger.error(f"Error generating accessibility snapshot: {str(e)}")
return {"status": "error", "message": str(e)}
def browser_console_messages(self, function_params: list = None) -> Dict[str, Any]:
"""
Retrieve JavaScript console messages (logs, warnings, errors) from the browser for debugging.
This function supports multiple parameter styles:
1. Standard style: no parameters
2. Nested function_params style:
function_params=[{"function_name": "browser_console_messages", "function_args": {}}]
Args:
function_params (list, optional): Nested function parameters
Returns:
Dict[str, Any]: The console messages including logs, warnings and errors
"""
# Check if browser is initialized
driver_check = self._check_driver_initialized()
if driver_check:
return driver_check
try:
logs = self._collect_browser_logs()
return {
"status": "success",
"console_messages": logs
}
except Exception as e:
logger.error(f"Error retrieving console messages: {str(e)}")
return {"status": "error", "message": str(e)}
def _collect_browser_logs(self) -> List[Dict[str, Any]]:
"""
Collect logs from both the browser driver and JavaScript console.
Returns:
List[Dict[str, Any]]: Combined logs from both sources
"""
logs = []
# Try to get browser logs if available
try:
browser_logs = self.driver.get_log('browser')
for log in browser_logs:
# Map browser log levels to standard levels
level = log.get("level", "").upper()
if level == "SEVERE":
level = "ERROR"
elif level == "INFO":
level = "LOG"
logs.append({
"level": level,
"message": log.get("message", ""),
"timestamp": log.get("timestamp", "")
})
except Exception as log_error:
# Browser logs might not be available in all drivers/configurations
logs.append({
"level": "WARNING",
"message": f"Could not retrieve browser logs: {str(log_error)}",
"timestamp": ""
})
# Try to execute JavaScript to get console message history
try:
# Add script to capture console messages if not already added
self.driver.execute_script("""
if (!window._consoleLogs) {
window._consoleLogs = [];
// Store original console methods
const originalConsole = {
log: console.log,
info: console.info,
warn: console.warn,
error: console.error,
debug: console.debug
};
// Helper function to add message with proper level
function addMessage(level, args) {
window._consoleLogs.push({
level: level.toUpperCase(),
message: Array.from(args).join(' '),
timestamp: new Date().toISOString()
});
}
// Override console methods to capture logs
console.log = function() {
addMessage('LOG', arguments);
originalConsole.log.apply(console, arguments);
};
console.info = function() {
addMessage('INFO', arguments);
originalConsole.info.apply(console, arguments);
};
console.warn = function() {
addMessage('WARN', arguments);
originalConsole.warn.apply(console, arguments);
};
console.error = function() {
addMessage('ERROR', arguments);
originalConsole.error.apply(console, arguments);
};
console.debug = function() {
addMessage('DEBUG', arguments);
originalConsole.debug.apply(console, arguments);
};
}
""")
# Wait a bit to ensure delayed messages are captured
time.sleep(2)
# Get the captured console logs
js_logs = self.driver.execute_script("return window._consoleLogs || [];")
# Add JS logs to our logs collection
for log in js_logs:
if log not in logs: # Avoid duplicates
logs.append(log)
except Exception as js_error:
logs.append({
"level": "WARNING",
"message": f"Could not retrieve JavaScript console logs: {str(js_error)}",
"timestamp": ""
})
return logs
def __del__(self):
"""
Destructor to automatically close the browser when the instance is destroyed.
"""
if hasattr(self, 'driver') and self.driver:
try:
self.driver.quit()
logger.info("Browser automatically closed on cleanup")
except Exception as e:
logger.warning(f"Error during automatic browser cleanup: {str(e)}")
class NavigateToUrlTool(Tool):
name: str = "navigate_to_url"
description: str = "Navigate to a URL and capture a snapshot of all page elements"
inputs: Dict[str, Dict[str, str]] = {
"url": {
"type": "string",
"description": "The complete URL (with https://) to navigate to"
},
"timeout": {
"type": "integer",
"description": "Custom timeout in seconds (default: 10)"
}
}
required: Optional[List[str]] = ["url"]
def __init__(self, browser_tool: BrowserBase = None):
super().__init__()
self.browser_tool = browser_tool
def __call__(self, url: str, timeout: int = None, function_params: list = None) -> Dict[str, Any]:
"""Navigate to URL using the BrowserBase instance."""
if not self.browser_tool:
raise RuntimeError("Browser tool instance not initialized")
try:
return self.browser_tool.navigate_to_url(url, timeout, function_params)
except Exception as e:
return {"status": "error", "message": f"Error navigating to URL: {str(e)}"}
class InputTextTool(Tool):
name: str = "input_text"
description: str = "Type text into a form field, search box, or other input element using a reference ID from a snapshot"
inputs: Dict[str, Dict[str, str]] = {
"element": {
"type": "string",
"description": "Human-readable description of the element (e.g., 'Search field', 'Username input')"
},
"ref": {
"type": "string",
"description": "Element ID from the page snapshot (e.g., 'e0', 'e1', 'e2'). Must refer to an editable element."
},
"text": {
"type": "string",
"description": "Text to input into the element"
},
"submit": {
"type": "boolean",
"description": "Press Enter after typing to submit forms (default: false)"
},
"slowly": {
"type": "boolean",
"description": "Type one character at a time to trigger JS events (default: true)"
}
}
required: Optional[List[str]] = ["element", "ref", "text"]
def __init__(self, browser_tool: BrowserBase = None):
super().__init__()
self.browser_tool = browser_tool
def __call__(self, element: str, ref: str, text: str, submit: bool = False, slowly: bool = True, function_params: list = None) -> Dict[str, Any]:
"""Input text using the BrowserBase instance."""
if not self.browser_tool:
raise RuntimeError("Browser tool instance not initialized")
try:
return self.browser_tool.input_text(element, ref, text, submit, slowly, function_params)
except Exception as e:
return {"status": "error", "message": f"Error inputting text: {str(e)}"}
class BrowserClickTool(Tool):
name: str = "browser_click"
description: str = "Click on a button, link, or other clickable element using a reference ID from a snapshot"
inputs: Dict[str, Dict[str, str]] = {
"element": {
"type": "string",
"description": "Human-readable description of what you're clicking (e.g., 'Login button', 'Next page link', 'Submit button')"
},
"ref": {
"type": "string",
"description": "Element ID from the page snapshot (e.g., 'e0', 'e1', 'e2'). You MUST get this ID from a previous snapshot's interactive_elements."
}
}
required: Optional[List[str]] = []
def __init__(self, browser_tool: BrowserBase = None):
super().__init__()
self.browser_tool = browser_tool
def __call__(self, element: str, ref: str, function_params: list = None) -> Dict[str, Any]:
"""Click element using the BrowserBase instance."""
if not self.browser_tool:
raise RuntimeError("Browser tool instance not initialized")
try:
return self.browser_tool.browser_click(element, ref, function_params)
except Exception as e:
return {"status": "error", "message": f"Error clicking element: {str(e)}"}
class BrowserSnapshotTool(Tool):
name: str = "browser_snapshot"
description: str = "Capture a fresh snapshot of the current page, including all elements"
inputs: Dict[str, Dict[str, str]] = {}
required: Optional[List[str]] = []
def __init__(self, browser_tool: BrowserBase = None):
super().__init__()
self.browser_tool = browser_tool
def __call__(self, function_params: list = None) -> Dict[str, Any]:
"""Take browser snapshot using the BrowserBase instance."""
if not self.browser_tool:
raise RuntimeError("Browser tool instance not initialized")
try:
return self.browser_tool.browser_snapshot(function_params)
except Exception as e:
return {"status": "error", "message": f"Error taking snapshot: {str(e)}"}
class BrowserConsoleMessagesTool(Tool):
name: str = "browser_console_messages"
description: str = "Retrieve JavaScript console messages (logs, warnings, errors) from the browser for debugging"
inputs: Dict[str, Dict[str, str]] = {}
required: Optional[List[str]] = []
def __init__(self, browser_tool: BrowserBase = None):
super().__init__()
self.browser_tool = browser_tool
def __call__(self, function_params: list = None) -> Dict[str, Any]:
"""Get console messages using the BrowserBase instance."""
if not self.browser_tool:
raise RuntimeError("Browser tool instance not initialized")
try:
return self.browser_tool.browser_console_messages(function_params)
except Exception as e:
return {"status": "error", "message": f"Error getting console messages: {str(e)}"}
class BrowserToolkit(Toolkit):
"""
Browser toolkit with auto-initialization and cleanup.
The browser is automatically initialized when any tool is first used,
and automatically closed when the toolkit instance is destroyed.
No explicit initialization or cleanup is required.
"""
def __init__(
self,
name: str = "BrowserToolkit",
browser_type: str = "chrome",
headless: bool = False,
timeout: int = 10,
**kwargs
):
# Create the shared browser tool instance
browser_tool = BrowserBase(
name="BrowserBase",
browser_type=browser_type,
headless=headless,
timeout=timeout,
**kwargs
)
# Create tools with the shared browser tool instance
# Note: Browser auto-initializes when first used and auto-closes when destroyed
tools = [
NavigateToUrlTool(browser_tool=browser_tool),
InputTextTool(browser_tool=browser_tool),
BrowserClickTool(browser_tool=browser_tool),
BrowserSnapshotTool(browser_tool=browser_tool),
BrowserConsoleMessagesTool(browser_tool=browser_tool)
]
# Initialize parent with tools
super().__init__(name=name, tools=tools)
# Store browser_tool as instance variable
self.browser_tool = browser_tool