🌐 Web-Use

Web-Use is an intelligent autonomous browsing agent, built to seamlessly navigate websites, interact with dynamic content, perform smart searches, download files, and adapt to ever-changing pages — all with minimal effort from you. Powered by advanced LLMs and the Chrome DevTools Protocol, it transforms complex web tasks into streamlined, automated workflows that boost productivity and save time.

• 🤖 Autonomous Web Navigation — Navigate websites, fill forms, and interact with dynamic content without manual intervention
• 🛠️ Multi-LLM Support — Works with Anthropic Claude, Google Gemini, OpenAI, Groq, Ollama, Cerebras, Mistral, and more
• 📸 Vision Capability — Understands visual content on pages for better decision-making
• 🔗 Web Model Context Protocol (WebMCP) — Discovers and uses custom tools exposed by websites, enabling context-aware interactions
• ⚡ Efficient Element Interaction — Indexed DOM elements for fast, accurate clicking and typing
• 📥 File Operations — Download files and upload content to forms
• 🔄 State Awareness — Maintains understanding of page state to avoid loops and recover from errors
• ⏱️ Intelligent Waiting — Handles loading states, animations, and user interactions (CAPTCHA, OTP)

Web-Use supports WebMCP, a protocol that allows websites to expose custom tools and capabilities directly to the agent. When visiting a website with WebMCP support:

• Auto-Discovery — The agent automatically detects available tools
• Dynamic Registration — Tools are added to the agent's toolkit on-the-fly
• Full Integration — WebMCP tools appear in the browser state with complete schema information
• Seamless Execution — Tools are called like built-in tools with proper parameter validation

If you visit a documentation site that supports WebMCP with a search_docs tool:

**WebMCP Tools Available:**
**search_docs** — Search documentation
  - `query` (string) [✓ required]
  - `limit` (integer) [○ optional]

The agent will automatically use this tool when relevant to the task.

Enable WebMCP support:

agent = Agent(
    config=config,
    llm=llm,
    use_web_mcp=True,  # Enable WebMCP discovery
    max_steps=100
)

• Python 3.11 or higher
• UV

Clone the repository:

git clone https://github.com/CursorTouch/Web-Use.git
cd Web-Use

Install dependencies:

uv sync

Launch Chrome with remote debugging:

chrome --remote-debugging-port=9222

Setting up the .env file:

GOOGLE_API_KEY="<API_KEY_HERE>"

Basic Setup:

from src.agent.browser.config import BrowserConfig
from src.providers.ollama import ChatOllama
from src.agent import Agent
from dotenv import load_dotenv

load_dotenv()

# Initialize LLM
llm = ChatOllama(model='qwen3.5:397b-cloud', temperature=0.5)

# Configure browser
config = BrowserConfig(
    browser='chrome',
    headless=False,
    use_system_profile=True
)

# Create agent with WebMCP support
agent = Agent(
    config=config,
    llm=llm,
    use_vision=False,
    use_web_mcp=True,  # Enable WebMCP for website tools
    max_steps=100
)

# Run agent
user_query = input('Enter your query: ')
agent.print_response(user_query)

Execute:

uv run main.py

config = BrowserConfig(
    browser='chrome',              # 'chrome' or 'edge'
    headless=False,                # Run in headless mode
    use_system_profile=True,       # Use real browser profile with auth
    user_data_dir='/path/to/profile',  # Custom profile directory
    cdp_port=9222,                 # Chrome DevTools Protocol port
    downloads_dir='/Downloads',    # Where to save files
    attach_to_existing=False,      # Connect to running browser
    update_cdp=False,              # Regenerate CDP protocol files
)

Prompt: I want to know the price details of the RTX 4060 laptop gpu from varrious sellers from amazon.in

Prompt: Make a twitter post about AI on X

Prompt: Can you play the trailer of GTA 6 on youtube

Prompt: Can you go to my github account and visit the Windows MCP

This project is licensed under MIT License - see the LICENSE file for details.

Contributions are welcome! Please see CONTRIBUTING for setup instructions and development guidelines.

Made with ❤️ by Jeomon George, Muhammad Yaseen

• Chrome DevTools Protocol
• LangGraph Examples
• vimGPT
• WebVoyager