Connects Claude to a full browser automation engine via async Python code execution. You get navigate, click, type, and screenshot primitives that run through Chrome DevTools Protocol, plus a persistent execution namespace so variables and state stick around between calls. The agent architecture has Claude write Python snippets that execute in a Jupyter-like environment rather than exposing fixed tools. Handles vision via screenshots, supports saved login sessions across conversations, and includes a CLI daemon mode for running automation scripts from the terminal. Reach for this when you need web scraping, form filling, or end-to-end testing where natural language describes the task but Python controls the browser.
claude mcp add --transport stdio billy-enrizky-openbrowser-ai uvx openbrowser-ai