Browser Kit MCP Server

Browser interaction in MCP Browser Kit starts with discovering connected browsers, windows, and tabs through getContext. Assistants can then inspect page text, enumerate interactive elements, or retrieve an element's HTML. Paginated snapshots let them continue through longer results. Interaction tools use readable element paths to click controls, enter values, and submit forms; separate scrolling tools move either the page or content inside a container. Tabs can also be opened or closed, and getSelection retrieves text the user has highlighted.

The extension build determines which fallback actions are available. MV2 supports tab screenshots, coordinate-based interaction, and JavaScript execution within the page when element or coordinate tools are insufficient. MV3 uses readable-element interaction rather than those fallbacks. For blocked steps such as CAPTCHAs, 2FA, or repeated automation failures, showHumanHint highlights a target and asks the user to act. Setup pairs an MCP client with a compatible browser extension, and open tabs need refreshing after the extension is enabled. The README identifies the project as alpha-stage, recommends a separate browser profile or instance to limit sensitive-data exposure, and warns that incoming requests currently have no authentication.