A new open-source tool called webctl is challenging the Model Context Protocol (MCP) approach to AI browser automation. Instead of relying on a server-controlled interface, webctl uses standard command-line interface (CLI) commands. This allows developers and AI agents to control browsers directly, offering a more transparent and scriptable alternative for automating web interactions.

The core frustration with MCP tools is the lack of control over context. MCP servers often flood AI context windows with excessive accessibility tree data and console logs. Webctl solves this by letting users filter exactly what enters the context. By piping output through standard Unix tools like grep or jq, users maintain strict control over data flow and efficiency.

Beyond filtering, the CLI architecture offers distinct advantages for debugging and scalability. Developers can cache snapshots to files, run parallel commands, and debug by executing the exact same commands as their agents. This stands in contrast to opaque server-side operations. Webctl essentially treats the browser as a standard command-line utility, fitting neatly into existing developer workflows.

To use webctl, you install it via pip and run a setup command to download a Chromium instance. The tool manages browser sessions as persistent daemons, allowing for stateful interactions like login cookies to survive across commands. It uses semantic ARIA role queries for element targeting, which remains stable even if a site's CSS changes.