langgraph-browser-agent 0.1.0

A web browser agent built with LangGraph

LangGraph Browser Agent

This project is a web browser agent built with LangGraph. It can navigate websites, fill out forms, and interact with web elements.

This library is built on top of LangGraph, a powerful framework for building agent applications, and leverages deepagents and langchain-mcp-adapters. It also supports two browser manipulation backends: playwright-mcp and BrowserMCP.

Demo

Getting Started

1. Prerequisites:

   This agent requires Node.js to be installed and the npx command to be available in your path.

   If you are using the browsermcp backend (the default), please read the documentation at https://docs.browsermcp.io/setup-extension to set up the browser extension.

2. Clone the repository:

   git clone https://github.com/your-username/langgraph-browser-agent.git
   cd langgraph-browser-agent
   

3. Install dependencies:

   uv pip install -e .
   

4. Set up your environment:

   This agent uses Ollama by default. Make sure you have Ollama installed and running. You can pull the default model with:

   ollama pull gpt-oss:20b
   

   You can also use other models by specifying them when creating the agent.

Usage

To use the browser agent in your own project, you can import the create_browser_agent function and use it in an async context:

import asyncio
from langgraph_browser_agent import create_browser_agent

OBJECTIVE = "Go to google.com and search for 'what is langgraph?'"

async def main() -> None:
    """Run the github example."""
    async with create_browser_agent(model="openai:gpt-5") as agent:
        final_result = await agent.ainvoke({"messages": [("human", OBJECTIVE)]})
        final_result["messages"][-1].pretty_print()

if __name__ == "__main__":
    asyncio.run(main())

You can also specify a different model when creating the agent:

async with create_browser_agent(model="ollama:qwen3:8b") as agent:
    # ...

Contributing

Contributions are welcome! Please open an issue or submit a pull request if you have any suggestions or improvements.

License

This project is licensed under the MIT License.