vercel-labs/agent-browser

What it solves

agent-browser is a fast, native Rust-based CLI tool designed to provide AI agents with a reliable way to automate and interact with web browsers. It simplifies the process of navigating pages, interacting with elements, and extracting data, while providing agent-friendly outputs like accessibility trees and annotated screenshots.

How it works

The tool operates as a CLI that controls a Chromium-based browser (via Chrome for Testing). It allows agents to perform a variety of actions—such as clicking, typing, and scrolling—using either traditional CSS selectors or semantic locators (ARIA roles, labels, and placeholders). To help AI agents "see" the page, it provides a snapshot command that generates an accessibility tree with unique references, allowing the agent to interact with elements by referring to those IDs (e.g., @e1) rather than complex selectors.

Who it’s for

It is primarily built for developers creating AI agents that need to browse the web, perform automated testing, or extract structured data from modern web applications.

Highlights

• Agent-Optimized Output: Generates accessibility trees and annotated screenshots to make page state easier for LLMs to parse.
• Semantic Locators: Supports finding elements by role, text, label, or placeholder, reducing reliance on fragile CSS selectors.
• Native Performance: Built with Rust for speed and efficiency.
• WebMCP Support: Includes experimental support for the Model Context Protocol (MCP) to discover and invoke page-provided tools.
• Comprehensive Browser Control: Full control over cookies, storage, network interception, viewport settings, and mouse movements (including human-like curved movement).
• Batch Execution: Supports executing multiple commands in a single call to reduce process startup overhead.