agent-browser

agent-browser is a browser automation CLI for AI agents that launches Chrome, uses ref-based snapshots, and offers broad command-line browser actions. Native binaries for macOS, Linux, and Windows.

AI Agent Development AI Developer Tools AI Testing & QA

Visit Website

Browser automation CLI for AI agents

agent-browser is a browser automation CLI for AI agents. It launches and controls Chrome through a native Rust daemon and uses compact text output to keep browser state readable in agent contexts.

The product is built around a ref-based snapshot workflow: the browser returns an accessibility tree with stable refs, and subsequent commands can interact with those refs directly. The docs position it for shell-driven agents and list support for navigation, forms, screenshots, network inspection, storage, tabs, frames, debugging, and other browser tasks.

Installation options include npm, npx, Homebrew, Cargo, and source builds, with native binaries available for macOS, Linux, and Windows. The site also documents configuration files, environment variables, and commands for persistent sessions, proxies, custom browser executables, and a browser automation MCP server.

Features

Compact ref-based snapshots

The snapshot command returns a compact accessibility tree with refs such as `@e1` and `@e2`, which is intended to reduce token usage and make element selection deterministic.

Broad command coverage

The CLI covers navigation, forms, screenshots, network, storage, files, tabs, frames, and debugging, with 50+ commands documented across the site.

Observable browser sessions

Built-in tools include video recording, streaming, debugging, profiler, and diffing, so browser sessions can be inspected without leaving the CLI.

Cross-platform native binaries

The docs highlight native support for macOS, Linux, and Windows, with Rust binaries rather than a wrapper around a browser extension.

Stateful browser sessions

Sessions can preserve profiles, auth state, cookies, storage, proxy settings, and security controls for longer-running workflows.

Advanced workflow support

The configuration and installation docs call out proxy support, init scripts, React and Web Vitals workflows, and Next.js + Vercel docs for app-specific automation.

Use Cases

AI-driven browser interaction
Use the CLI to open a page, snapshot it, and click or fill fields by ref when an AI agent needs deterministic browser control from the terminal.
Long-running authenticated workflows
Use persistent sessions, profiles, cookies, and storage when you need the browser to stay logged in across a longer workflow or repeated runs.
Debugging and visual inspection
Use screenshots, video recording, streaming, profiler, and diffing when you need to inspect rendering changes or capture browser behavior for debugging.
Controlled environments and CI
Use proxy settings, custom launch arguments, and install-time dependency helpers when automating in CI, a devcontainer, or behind enterprise network constraints.
Agent-assisted development
Use the documented native integrations and shell-command workflow when pairing with coding agents such as Claude Code, Cursor, GitHub Copilot, or OpenAI Codex.

Pros and Cons

Pros

Compact text output is designed to use fewer tokens than full DOM or JSON output.
Snapshots expose element refs that make interactions deterministic across follow-up commands.
The CLI includes a wide set of documented browser actions, from navigation and forms to screenshots, network, storage, and debugging.
Native binaries are available for macOS, Linux, and Windows.
Configuration supports persistent defaults, environment overrides, and project-level settings.

Cons

The pricing page does not currently provide a stable published plan summary in the collected evidence.
Some integrations and provider details are documented only partially in the collected sources, so the supported ecosystem is broader than the audited evidence can confirm.

FAQ

What is agent-browser used for?

It is a browser automation CLI designed for AI agents. The docs describe commands for opening pages, taking snapshots, clicking by refs, filling forms, capturing screenshots, and connecting to browsers through CDP.

How do I install agent-browser?

The installation docs show global install, project install, Homebrew on macOS, Cargo, and npx quick start options. Linux users can also install required system dependencies with `agent-browser install --with-deps`.

How does the basic workflow work?

The command reference shows a shell-oriented workflow: open a page, take a snapshot to get element refs, and then interact using those refs with commands like click, fill, type, and screenshot.

What kinds of commands does it support?

The docs list `snapshot`, `screenshot`, `pdf`, `read`, `stream`, `connect`, and `mcp` among the available commands, along with navigation, form, tab, file, and debugging actions.

Which platforms and agents does it support?

The site documents native binaries for macOS, Linux, and Windows, and says it works with agents that can run shell commands, including Claude Code, Cursor, GitHub Copilot, OpenAI Codex, and Gemini.