44.1K ·GitHub
What is this?
agent-browser is a browser automation CLI purpose-built for AI agents, shipped as a fast native Rust binary that drives Chrome for scripted or agentic web interaction.
agent-browser is a browser automation CLI purpose-built for AI agents, shipped as a fast native Rust binary that drives Chrome for scripted or agentic web interaction.Is this a library I import, or a standalone CLI/MCP tool?It is a standalone command-line binary (not a code library) that you invoke directly or wire up as an MCP tool for agents; it drives a managed Chrome instance via a background daemon.
It is a standalone command-line binary (not a code library) that you invoke directly or wire up as an MCP tool for agents; it drives a managed Chrome instance via a background daemon.How do I install and start using it?Install the native binary globally (recommended), via Homebrew on macOS, via Cargo, or build from source, then run its install command to fetch Chrome for Testing before issuing CLI commands like click, fill, or navigate.
Install the native binary globally (recommended), via Homebrew on macOS, via Cargo, or build from source, then run its install command to fetch Chrome for Testing before issuing CLI commands like click, fill, or navigate.Does it need Chrome, Node, or Rust installed first?It needs Chrome, which it can auto-download via 'agent-browser install', or it detects an existing Chrome, Brave, Playwright, or Puppeteer install; Node.js 24+/pnpm 11+ and Rust are only needed if you build from source.
It needs Chrome, which it can auto-download via 'agent-browser install', or it detects an existing Chrome, Brave, Playwright, or Puppeteer install; Node.js 24+/pnpm 11+ and Rust are only needed if you build from source.Are there any gotchas with waiting or accessibility checks?The networkidle wait condition can fail to resolve on pages using SSE, WebSockets, polling, or long-polling, and accessibility audits are not available on Safari or iOS WebDriver sessions, only on CDP-based Chrome.
The networkidle wait condition can fail to resolve on pages using SSE, WebSockets, polling, or long-polling, and accessibility audits are not available on Safari or iOS WebDriver sessions, only on CDP-based Chrome.Does this work on macOS, Windows, and Linux the same way?Yes, it runs on macOS, Windows, and Linux, but on Linux you must install extra system libraries so Chrome can launch correctly.
Yes, it runs on macOS, Windows, and Linux, but on Linux you must install extra system libraries so Chrome can launch correctly.What is WebMCP and why does it matter here?WebMCP is an experimental feature, enabled by default, that lets agent-browser discover and surface tools exposed by a page itself, announcing them automatically in CLI/MCP responses so an agent can call page-native actions.
WebMCP is an experimental feature, enabled by default, that lets agent-browser discover and surface tools exposed by a page itself, announcing them automatically in CLI/MCP responses so an agent can call page-native actions.Can it run accessibility checks on a page?Yes, it embeds the axe-core engine to run full accessibility audits offline, returning violations with impact, rule id, selectors, and fix guidance, or structured JSON for automation.
Yes, it embeds the axe-core engine to run full accessibility audits offline, returning violations with impact, rule id, selectors, and fix guidance, or structured JSON for automation.Does it offer anything for React apps specifically?Yes, launching with --enable react-devtools embeds the React DevTools hook so you can introspect React component state, plus framework-agnostic Web Vitals metrics work on any app.
Yes, launching with --enable react-devtools embeds the React DevTools hook so you can introspect React component state, plus framework-agnostic Web Vitals metrics work on any app.Can I batch multiple browser actions into one call?Yes, batch execution lets you pass multiple commands as quoted arguments or piped JSON via stdin, avoiding per-command process startup overhead for multi-step workflows.
Yes, batch execution lets you pass multiple commands as quoted arguments or piped JSON via stdin, avoiding per-command process startup overhead for multi-step workflows.What are similar projects?bb-browser, oc, playwright, agent-desktop, XActions, gemini-notebook-mcp-cli, Agent-Reach, playwright-cli
bb-browser, oc, playwright, agent-desktop, XActions, gemini-notebook-mcp-cli, Agent-Reach, playwright-cliWhich topics is it in?chrome, accessibility, web-scraping, homebrew, browser-automation, model-context-protocol, cli-tool, testing, command-line, ai-agent, rust, react
chrome, accessibility, web-scraping, homebrew, browser-automation, model-context-protocol, cli-tool, testing, command-line, ai-agent, rust, react