Browserbase MCP Server
Cloud browser automation for LLMs using Browserbase and Stagehand tools.
Data last scanned today · Reviewed yesterday
Overview
This server bridges Large Language Models with Browserbase and Stagehand, enabling AI agents to browse, interact with, and extract data from websites. It provides six specific tools, including navigation, observation of actionable elements, and automated browser actions for complex web workflows. Developers can deploy this via STDIO or connect to the hosted SHTTP version provided by the vendor. The functionality mirrors the hosted Browserbase MCP server, allowing for direct integration into AI-powered IDEs or chat interfaces that support the Model Context Protocol.
Our verdict
With a security score of 38/100, this server is classified as high-risk. The primary concern is the total lack of an authentication method, meaning any local configuration access enables full control over the browser session. Furthermore, the absence of a read-only mode means the server has full write access by default. It is suitable for controlled, sandboxed development environments but should not be exposed to untrusted users or network environments without implementing external security layers.
- Supports browser automation via Stagehand and Browserbase
- Six dedicated tools for web navigation and data extraction
- Small dependency footprint with only 6 direct dependencies
- Active development with 15 contributors
- No authentication mechanism implemented
- Lacks a read-only mode for restricted browser access
- Community-maintained rather than official vendor-managed
- Not yet undergone editorial security review
Setup
{
"mcpServers": {
"mcp-server-browserbase": {
"args": [
"-y",
"@browserbasehq/mcp"
],
"command": "npx"
}
}
}Paste into ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows), then restart Claude Desktop. Some servers require API keys or tokens as environment variables — check the project's README and add an "env" object if needed.
- No read-only mode — running this grants full read/write access.
- No built-in authentication — anyone with access to this config can invoke every tool it exposes.
Tools
| Tool | Description | Risk |
|---|---|---|
| start | Create or reuse a Browserbase session | medium |
| end | Close the current Browserbase session | medium |
| navigate | Navigate to a URL | medium |
| act | Perform an action on the page | medium |
| observe | Observe actionable elements on the page | low |
| extract | Extract data from the page | low |
Compatibility
| Client | Local | Docker | Remote | Read-only |
|---|---|---|---|---|
| ChatGPT | ||||
| Claude Desktop | ||||
| Cursor | ||||
| VS Code | ||||
| Windsurf |
Frequently asked questions
›What is the recommended way to get started with the Browserbase MCP server?
The project recommends using the hosted version at https://mcp.browserbase.com/mcp, as it is the easiest to set up and provides LLM costs for the Gemini model.
›What API keys are required to run the self-hosted version?
You must provide a BROWSERBASE_API_KEY, a BROWSERBASE_PROJECT_ID, and a GEMINI_API_KEY in your environment configuration.
›Can I use a different LLM model instead of the default?
Yes, you can specify a different model by adding the --modelName flag to your configuration arguments along with its respective API key.
›Are there specific features available only for Scale Plan users?
Yes, the --verified flag, which enables Browserbase Verified Identity, is restricted to Scale Plan users.
›How can I configure the viewport size of the browser session?
You can adjust the dimensions using the --browserWidth and --browserHeight flags, which default to 1024 and 768 pixels respectively.
Alternatives
Changelog
- Breakingv3.0.0
2026-03-31
Badge
Maintain this server? Add the live badge to your README.