Browser Use

An MCP server enabling AI agents to control web browsers via browser-use, supporting SSE and stdio protocols, with VNC streaming for real-time observation.

Connections & Capabilities

Connects To

GitHubDiscordDockerPlaywright

Capabilities

readwriteexec

Quickstart

Config

{
  "mcpServers": {
    "browser-use-mcp-server": {
      "url": "http://localhost:8000/sse"
    }
  }
}

Exposed MCP Tools (5)

moderate

open

Opens a specified URL in the browser.

Navigating to malicious sites could be harmful.

safe

return_the_top_ranked_article

Extracts the top ranked article from a webpage.

Read-only operation; no modification of data.

moderate

click

Clicks on a specified element within the browser.

Clicking on unintended elements could lead to unwanted actions.

moderate

type

Types text into a specified element within the browser.

Typing into unintended fields could lead to unwanted data submission.

safe

get_content

Retrieves the content of a specified element within the browser.

Read-only operation; no modification of data.

Safety Assessment

The server's safety depends on proper configuration and secure handling of the OpenAI API key. It's relatively safe if used in a controlled environment with a strong VNC password. Risks increase if exposed to the internet or untrusted agents.

Requires explicit configuration to connect to clients.
Browser automation is confined to the browser environment.
Supports running within Docker for isolation.
VNC password can be customized for added security.

Can perform arbitrary actions within a web browser.
Requires an OpenAI API key, which needs secure handling.
If misconfigured, could expose browser control to unauthorized agents.
Execution of arbitrary javascript in the browser context.