agent-browser is a browser automation CLI for AI agents built as a fast native Rust binary.
The tool addresses the need for AI systems to interact with web browsers programmatically. It provides a command-line interface that allows agents to perform actions like clicking elements, filling forms, taking screenshots, and reading page content. The approach centers on semantic locators that identify elements by their accessibility role and name rather than brittle CSS selectors, reducing failures when page structure changes. It includes built-in handling for common blocking scenarios, such as dismissing consent banners before retrying clicks, and offers multiple content extraction modes optimized for AI consumption, including markdown conversion and llms.txt file discovery.
The tool suits projects where AI agents need reliable web interaction without the overhead of Node.js or Playwright. It works well for automation tasks that benefit from semantic element selection and for scenarios where agents must navigate authenticated sessions or client-side-rendered content. The project detects existing Chrome, Brave, Playwright, and Puppeteer installations automatically, reducing setup friction. Installation options include global binary installation, local npm dependency pinning, Homebrew, and Cargo, making it adaptable to different workflow preferences.
Development activity shows consistent maintenance with regular updates to the update mechanism itself, suggesting active refinement of the installation and upgrade experience. The project maintains comprehensive documentation of command options and behaviors, including detailed guidance on edge cases like scrollbar visibility and element occlusion handling. Work spans both the core Rust binary and integration with multiple package managers, indicating attention to developer experience across different environments.
