home / mcp / mcp browser use server

MCP Browser Use Server

FastAPI server implementing MCP protocol Browser automation via browser-use library.

Installation

Add the following to your MCP client configuration file.

Configuration

{
  "mcpServers": {
    "jovanipink-mcp-browser-use": {
      "command": "uvx",
      "args": [
        "mcp-browser-use"
      ],
      "env": {
        "MCP_MODEL_NAME": "claude-3-5-sonnet-20241022",
        "MCP_MODEL_PROVIDER": "anthropic"
      }
    }
  }
}

You can run the MCP Browser Use Server to automate browser interactions through browser-use, orchestrating a browser session and returning results to your MCP client. This server provides a single tool, run_browser_agent, to manage a browser session end-to-end and integrate with MCP-compatible clients like Claude Desktop.

How to use

Start by launching the MCP Browser Use Server locally and connect your MCP client to it. The server exposes a single tool, run_browser_agent, which starts a browser session, runs the browser-use agent, and returns the final result to your client. You can test and observe tool calls in real time using the MCP Inspector for debugging and visibility into prompts, actions, and responses.

To integrate with Claude Desktop or another MCP-compatible client, configure the MCP server entry with the provided runtime command and environment settings. You will use the local stdio flow, where your client communicates via the standard input/output channels established by the runtime.

How to install

Prerequisites you need to prepare before installing and running the server:

Python 3.11 or newer
Google Chrome or Chromium for local browser automation
uv for dependency management (recommended)
Optional: Claude Desktop or another MCP-compatible client for testing

Step-by-step installation and startup:

uv sync

uv run mcp-browser-use

# Copy sample.env to .env and configure your providers
# or export variables in your environment as needed

# Optional: test with the MCP Inspector for debugging
npx @modelcontextprotocol/inspector uv --directory /path/to/project run mcp-browser-use

Configuration and running notes

This server uses environment variables to control model providers, model names, and browser behavior. The key options include provider and model name, temperature, and flags for vision and action limits. Keep credentials secure by using a secrets manager or a local .env file that is not committed to source control.

When starting with Claude Desktop, the MCP server is exposed as a local stdio service. Use the following runtime configuration to register the MCP server in Claude Desktop, ensuring the environment variables match your chosen model provider and model.

{
  "mcpServers": {
    "mcp_server_browser_use": {
      "command": "uvx",
      "args": ["mcp-browser-use"],
      "env": {
        "MCP_MODEL_PROVIDER": "anthropic",
        "MCP_MODEL_NAME": "claude-3-5-sonnet-20241022"
      }
    }
  }
}

Security considerations

Running a full browser instance remotely can grant broad access to the host. Ensure you review security considerations and restrict access to trusted clients and environments. Use strong authentication for MCP client connections and keep your browser automation environment isolated from sensitive networks.

Troubleshooting and debugging

If you need interactive debugging, use the MCP Inspector to watch tool calls and responses in real time. The inspector provides a URL you can open in a browser to observe the lifecycle of a run, including prompts, actions, and results.

Notes on tools and endpoints

This server exposes a single MCP tool named run_browser_agent which orchestrates a browser session and delegates to the browser-use agent. You interact with this tool through your MCP client, sending the appropriate prompts and actions and receiving the final result.

Available tools

run_browser_agent

Orchestrates a browser session, runs the browser-use agent, and returns the final result to the MCP client.