agent-browser

What is agent-browser?

agent-browser is a browser automation CLI designed for AI coding agents rather than for people writing test scripts. You install it with npm or Homebrew, download Chrome once, then drive pages with shell commands such as open, snapshot, click, fill and screenshot. The snapshot command prints a compact accessibility tree in which every element has a ref such as @e1, so an agent can target an exact element without sending the full DOM; the docs estimate roughly 200 to 400 tokens, compared with 3,000 to 5,000. A Rust CLI talks to a native daemon that controls Chrome through the DevTools Protocol, with sessions, profiles, cookies, proxies and an idle timeout. Built-in extras include video recording, streaming, a local observability dashboard, network control and an MCP stdio server. It works with Claude Code, Cursor, GitHub Copilot, OpenAI Codex and any agent that can run shell commands. No pricing is listed, and it runs on macOS, Linux and Windows.

Use cases

• Automate browsers from coding agents
• Run exploratory tests on web apps
• Scrape pages into agent-readable text
• Record network traffic into a client
• Automate Electron apps
• Test Vercel deployments behind auth

Who it's for

• AI agent developers
• Developers using coding agents
• QA engineers

Why use agent-browser

• Short text snapshots keep an agent's context window free for the actual task.
• Refs point to exact elements, so actions are repeatable rather than guessed from selectors.
• Any agent that can run shell commands can use it, with no custom integration.
• Skills served by the CLI match the installed version, so instructions do not drift.
• A local dashboard lets you watch what an agent is doing in the browser.

agent-browser features

• Compact accessibility-tree snapshots with element refs
• 50+ commands for navigation, forms, files, tabs and network
• Native Rust CLI and daemon over Chrome DevTools Protocol
• Screenshots with annotation and change detection
• Sessions, profiles, cookies, auth state and proxy support
• Local observability dashboard with live viewport
• Video recording, streaming and profiler tools
• Built-in MCP stdio server and WebMCP tool discovery
• Skills served by the CLI for agents such as Claude Code and Cursor
• Cloud browser providers including AgentCore and Browserbase

agent-browser integrations

• Claude Code
• Cursor
• GitHub Copilot
• OpenAI Codex
• Google Gemini
• opencode
• Vercel Sandbox
• Vercel AI Gateway
• AWS Bedrock AgentCore
• Browserbase
• Browserless
• Kernel

How to get started with agent-browser

1. Install with npm install -g agent-browser or brew install agent-browser, then run agent-browser install.
2. Open a page with agent-browser open and capture a snapshot with refs.
3. Act on elements by ref, for example agent-browser click @e2, then take a screenshot.
4. Optionally add the discovery skill with npx skills add vercel-labs/agent-browser.
5. Start the dashboard to watch sessions live, and close the browser when done.

agent-browser limitations

• The daemon closes its headless browser after one hour without commands unless configured otherwise
• Clicks fail when another element covers the target or the element is hidden or disabled
• Native binaries are listed for macOS ARM64 and x64, Linux ARM64 and x64, and Windows x64 only
• The dashboard AI chat requires an AI Gateway API key

agent-browser pricing

agent-browser is open source, so you can use and self-host it for free.

Prices can change and may vary by country or exclude tax. Check the official pricing page.

agent-browser alternatives

The main agent-browser alternatives listed on Debuly are Playwright, Browser Use, Stagehand, Browserbase, Hyperbrowser and Runloop. The table compares pricing and platforms; order is A–Z.

1. 

   Browser Use

   Cloud browsers and hosted web agents for AI, plus an MIT-licensed open-source browser agent, billed pay-as-you-go from $0.02 per hour.

   Developer platformsAutomation and agentsCoding
   PaidVisit Browser Use (official site)
2. 

   Browserbase

   Cloud platform of headless browsers, search and fetch APIs and agent runtime for building AI agents that browse and act on the web.

   Developer platformsAutomation and agentsCoding
   FreemiumVisit Browserbase (official site)
3. 

   Hyperbrowser

   Cloud infrastructure that gives AI agents hosted Chrome sessions, isolated sandboxes, and web fetch, crawl and search APIs.

   Developer platformsAutomation and agentsCoding
   FreemiumVisit Hyperbrowser (official site)
4. 

   Playwright

   Open-source Microsoft framework for end-to-end web testing and browser automation across Chromium, Firefox and WebKit, with CLI and MCP for agents.

   Developer platformsCodingAutomation and agents
   Open sourceVisit Playwright (official site)
5. 

   Runloop

   Sandboxed devboxes, blueprints and snapshots for building and running AI software engineering agents.

   Developer platformsCodingAutomation and agents
   FreemiumVisit Runloop (official site)
6. 

   Stagehand

   Open-source SDK from Browserbase for building browser agents with act, extract and observe calls in TypeScript, Python or Go.

   Developer platformsAutomation and agentsCoding
   Open sourceVisit Stagehand (official site)

agent-browser FAQ

Sources & verification

Facts on this page come from the official sources below, each with the date we read it. Written by Debuly; the vendor did not review this page.

Image credits

• Logo: from agent-browser.dev · fetched 23 Oct 2026 · Vendor trademark, used only to identify the product; removed on request
• Cover: none collected yet, so a generated tile is shown.

Spot an error? Report an issue with this listing. We also remove a logo or image on the owner's request.