MCP Browser Agent

You gain browser automation capabilities through an MCP server that lets you control a Chrome browser from an MCP client. It provides state management, planning, and action execution to automate web tasks in a structured, repeatable way.

How to use

You connect to the MCP Browser Agent from your MCP client to retrieve the current browser planning state and to execute planned actions in the browser. Use the two main tools the server exposes: get_planner_state to inspect the current browser state and planning context, and execute_actions to perform the planned actions inside the browser. You can configure multiple browser contexts, enable interactive element handling, and log debugging information to help you refine automation tasks. Ensure Google Chrome is installed and available on your system, and that the agent is running in a configured environment.

How to install

# Prerequisites
- macOS (darwin 24.2.0)
- Python 3.12 or higher
- `uv` package manager
- Google Chrome

# Installation via Smithery
npx -y @smithery/cli install @ashley-ha/mcp-manus --client claude

# Manual Installation
1. Clone the repository:
```bash
git clone <repository-url>
cd mcp

2. Set up the Python environment using uv:

uv venv
source .venv/bin/activate
uv sync
`

Additional sections

Configuration and runtime behavior are designed to be straightforward. The MCP server uses a stdio entry point to start the agent locally, and you can point Claude Desktop or another MCP client to it for browser automation tasks.

Configuration

{
  "mcpServers": {
    "browser-use": {
      "command": "uv",
      "args": [
        "--directory",
        "/ABSOLUTE/PATH/TO/mcp",
        "run",
        "browser-use.py"
      ]
    }
  }
}

Notes on browser and environment

The agent is configured to use Google Chrome with the following default settings: non-headless mode for development, window size 1280x1100, security features may be relaxed for testing, and recordings saved under ./tmp/recordings. Adjust these settings as needed for your environment and security requirements.

Development and logging

Logging is implemented with Python’s built-in facilities. Logs go to stderr with a standard format. The root logger operates at INFO level, while third-party loggers run at WARNING by default.

Project structure and tools

Key components include the main entry point browser-use.py which implements the MCP server, and a tmp/recordings directory for browser session recordings. Dependencies are managed via uv.

Contributing and license

Contributions are welcome. This project is licensed under the MIT License, and you should review the LICENSE file for full terms.

Available tools

get_planner_state

Retrieves the current browser state and planning context from the MCP server so you can understand the planned actions and the current browser context.

execute_actions

Executes the actions that have been planned and prepared for the browser, carrying out automated interactions within the Chrome session.