Give your AI agent a real browser it can control from the command line. Agent Browser is a fast, native Rust CLI that launches Chrome, takes accessibility snapshots, and executes precise actions so your agent can navigate, click, fill forms, and extract data reliably.
•What it does
Launches and controls Chrome via Chrome DevTools Protocol with a persistent daemon for low‑latency commands.
AI‑friendly snapshots: agent-browser snapshot returns an accessibility tree with stable refs (e.g. @e1, @e2) so agents can click/fill by semantic reference instead of brittle CSS selectors.
Full interaction set: click, double‑click, type, fill, press keys, hover, drag & drop, file upload, scroll, select, check/uncheck, and more.
Screenshots, PDFs, and video: capture pages, full‑page screenshots, annotated screenshots, PDFs, and record WebM/MP4 sessions with optional cursor overlay.
Network & state control: route/block/mock requests, inspect traffic, record HAR, manage cookies, localStorage/sessionStorage, and save/load auth state.
Tabs, frames, dialogs: multi‑tab workflows with stable t1, t2 handles and labels; iframe switching; explicit dialog (alert/confirm/prompt) handling.
Advanced dev tools: React DevTools integration (tree, inspect, render profiling, Suspense), Web Vitals metrics, axe‑core accessibility audits, tracing/profiling, console/error inspection.
•Built for AI agents
MCP server: agent-browser mcp exposes typed tools over Model Context Protocol so Claude Code, Cursor, and other MCP clients can drive the browser with approval prompts and structured arguments.
Skills system: agent-browser skills serves up‑to‑date skill content (e.g. protected Vercel deployments) directly from the installed CLI version.
Chat mode: agent-browser chat provides natural‑language, single‑shot or REPL‑style browser control.
•Traditional selectors also work:
Authentication & sessions
Reuse your existing Chrome profile (--profile), persist cookies/storage with --session --restore, or import auth from a running Chrome via --auto-connect + state save.
Run multiple isolated browsers with --session agent1, --session agent2, each with its own cookies, storage, and history.
• Who it’s for
AI coding assistants and research agents that need reliable web access
Developers building browser automation, scraping, or testing pipelines
Teams that want MCP‑exposed, typed browser tools for LLM clients
No comments or reviews, maybe you want to be first?