Browser Use MCP Server
Automate web browsing tasks by enabling AI agents to click, type, and navigate pages.
Data last scanned today · Reviewed yesterday
Overview
This server wraps the browser-use library, providing AI models the capability to control a real web browser. It automates common interactive tasks like filling forms, clicking elements, and navigating complex websites to extract data or perform research operations. Built to overcome common stdio timeout issues, the server utilizes an HTTP transport mechanism to support long-running browser tasks that can last over a minute. It also includes a dedicated web UI for monitoring task progress, execution logs, and overall server health.
Our verdict
This tool carries a high risk level with a security score of 24/100, primarily due to the lack of any authentication mechanism, which allows anyone with access to the configuration to control the server. Additionally, it lacks a read-only mode, meaning all installations have full write access to the browser environment. Given these factors and the community-maintained nature of the project, it should only be used by developers in isolated, local testing environments rather than exposed or shared machines.
- Uses HTTP transport to prevent long-running task timeouts
- Includes a built-in web UI for monitoring live automation
- Zero direct dependencies minimize the supply-chain surface
- No authentication method leaves the server open to local access
- Full write access is granted by default with no read-only mode
- Community-maintained project not backed by an official vendor
Tools
| Tool | Description | Risk |
|---|---|---|
| run_browser_agent | Execute browser automation tasks | high |
| run_deep_research | Multi-search research with synthesis | high |
| skill_list | List learned skills | low |
| skill_get | Get skill definition | low |
| skill_delete | Delete a skill | medium |
| health_check | Server status and running tasks | low |
| task_list | Query task history | low |
| task_get | Get full task details | low |
Compatibility
| Client | Local | Docker | Remote | Read-only |
|---|---|---|---|---|
| ChatGPT | ||||
| Claude Desktop | ||||
| Cursor | ||||
| VS Code | ||||
| Windsurf |
Frequently asked questions
›Why does this server use HTTP instead of stdio?
The server uses HTTP transport to prevent connection timeouts during long-running browser automation tasks, which typically take between 30 and 120 seconds to complete.
›How can I monitor browser tasks while they are running?
You can access the built-in web dashboard at http://localhost:8383/dashboard to view real-time task progress, execution logs, and server health.
›Does this server require an LLM to function?
Yes, the browser agent requires an LLM to decide on actions. You must provide an API key for your chosen provider via environment variables or the configuration file.
›Can I use an existing Chrome browser profile with this server?
Yes, you can configure the browser to use a specific user data directory to maintain persistent logins and cookies by setting the 'browser.user_data_dir' configuration option.
›How do I connect this server to an MCP client that does not support HTTP transport?
You can use 'mcp-remote' as a proxy by configuring your MCP client to execute 'npx' with the 'mcp-remote' argument pointing to the server's local HTTP URL.
›What is the order of precedence for server configuration settings?
Configuration settings are applied in the following priority: environment variables take precedence, followed by the configuration file, and finally the default values.
Alternatives
Badge
Maintain this server? Add the live badge to your README.