Agent Browser

Agent Browser is Vercel Labs’ open-source browser automation CLI designed for AI agents, with 44,300+ stars on GitHub.

Overview

Agents that need the web usually get a heavyweight framework or a pile of screenshots to squint at. Agent Browser takes a leaner approach: a fast Rust command-line tool that drives headless Chrome through compact, scriptable commands. Its key idea is the snapshot: ask for the page’s accessibility tree and every interactive element comes back with a stable reference, so the agent can click, fill and select by reference instead of parsing raw HTML or guessing coordinates — an interaction model that uses far fewer tokens and breaks far less often. Around that core it provides the practical tooling automation needs: screenshots and PDFs, multiple sessions, cookie and storage management, and waiting strategies for dynamic pages. Agent Browser is built by Vercel Labs, documented at agent-browser.dev, and released under the Apache License 2.0; it slots into agent frameworks, CI jobs and MCP-style workflows as the browsing hand of an AI system.

Key Features

• Built for agents – a CLI designed around how LLMs actually interact with pages.
• Snapshot with element refs – accessibility-tree snapshots with stable references for clicks and typing.
• Fast Rust core – quick startup and snappy command execution for long automation runs.
• Screenshots and PDFs – capture pages visually whenever the agent needs proof.
• Sessions and cookies – keep logins and state across steps and runs.
• Dynamic-page handling – waits and retries tuned for modern JavaScript-heavy sites.
• Framework friendly – call it from agent frameworks, scripts or CI pipelines.

💳 Pricing

Agent Browser is free and open source under the Apache License 2.0. There is no hosted tier to buy — you run the browser on your own machines.

Who Is It For

Agent developers and automation engineers who need dependable, low-token web automation their models can drive directly from the terminal or code.