Set up Browserbase MCP Server

What the Browserbase MCP Server does

The Browserbase MCP Server connects AI language models to cloud browser infrastructure, giving AI agents access to real, automated browser sessions. It is built on the Model Context Protocol, an open standard for handing structured tools and context to AI models, and on Stagehand, a framework for reliable AI-driven browser control. Instead of launching a browser locally, the server opens a session inside Browserbase's cloud infrastructure, navigates to web pages, observes actionable elements, performs actions, and extracts data from pages. This makes tasks such as research, form filling, screenshots, or content extraction possible directly from an AI chat, without a person having to operate a browser themselves.

Important note on the project's status

According to the GitHub repository, the original self-hostable server code has since been archived and is no longer actively maintained. The provider explicitly labels the repository as a historical reference implementation and states it should not be interpreted as representative of Browserbase's current production service. Anyone looking to use the service in production today should therefore check the hosted variant described in the official Browserbase documentation first, since that stays current, while the archived open-source code remains a learning resource and a starting point for custom modifications.

Tools and how it works

Per the documentation, the server exposes six core tools matching the scope of the hosted variant. A session can be started or an existing one reused, a running session can be closed, the browser can navigate to a URL, an action can be performed on the page, actionable elements on the page can be observed, and data can be extracted from the current page content. For communication, the server supports two transports: STDIO for locally running processes and Streamable HTTP for connecting to the hosted infrastructure. In the hosted variant, Browserbase states it also covers the cost of the underlying language model, naming Gemini as the best-performing model for Stagehand. Anyone who prefers a different model can configure it through an additional parameter along with a matching API key.

Prerequisites, security, and setup

The self-hosted variant requires Node.js and npm plus a Browserbase account with an API key and project ID; alternatively, the server can run via Docker using a locally built container image. Credentials such as API keys belong in environment variables of the respective MCP client configuration and should never be stored in plain text inside source code or chat transcripts. Because the server visits web pages and performs actions on a user's behalf, its use should always be restricted to trusted target sites and clearly scoped tasks to avoid accidental purchases, data leakage, or violations of a site's terms of use. In practice, the service is well suited to automated research, monitoring web content, or repetitive form-filling tasks, but it hits limits once target sites deploy aggressive bot detection or require complex, multi-step authentication. Because the code has been archived, it is worth checking the current official documentation before any production use to confirm that configuration and tool scope still match the actual offering.