---
title: "agent-browser skill: browser automation CLI for AI agents"
canonical_url: https://cookiesforai.app/skills/vercel-labs/agent-browser
last_updated: 2026-10-02T01:58:30+02:00
type: skill
summary: "agent-browser is the agent skill for Vercel Labs' browser automation CLI: a native Rust tool that lets an agent open pages, read accessibility snapshots and click, fill and screenshot by element reference."
install: "npx skills add vercel-labs/agent-browser --skill agent-browser"
related:
  - "https://cookiesforai.app/skills/firecrawl/firecrawl"
---

# agent-browser

agent-browser is the agent skill for Vercel Labs' browser automation CLI: a native Rust tool that lets an agent open pages, read accessibility snapshots and click, fill and screenshot by element reference.

| Fact | Value |
|---|---|
| Publisher | Vercel (official, own GitHub account) |
| Source | https://github.com/vercel-labs/agent-browser/tree/HEAD/skills/agent-browser |
| License | Apache-2.0 |
| Repository stars | 43,444 on 2026-10-02 |
| Skill name | agent-browser |
| Category | Browser automation and web research |
| Skill last changed | 2026-08-24 |
| Page updated | 2026-10-02 |
| Facts checked | 2026-10-02 |

## Install

Install the skill with the skills CLI:

```
npx skills add vercel-labs/agent-browser --skill agent-browser
```

Install the agent-browser CLI, which the skill drives:

```
npm install -g agent-browser
```

Download Chrome for Testing the first time:

```
agent-browser install
```

## Ask your coding agent

A skill can include scripts that run on your computer, so read its source first: https://github.com/vercel-labs/agent-browser/tree/HEAD/skills/agent-browser

```text
Install the agent skill agent-browser from github.com/vercel-labs/agent-browser.
```

## What it does

agent-browser is the skill that ships with Vercel Labs' browser CLI of the same name. The CLI drives Chrome or Chromium over the DevTools protocol without Playwright or Puppeteer. The agent opens a page, takes an accessibility-tree snapshot, and acts on short references such as @e2 to click, fill, read text or take screenshots, then snapshots again after the page changes.

The SKILL.md in the repository is deliberately thin. It points the agent to agent-browser skills get core, which prints workflows and troubleshooting from the installed CLI, so the instructions always match the version on your machine. Specialized guides load the same way for Electron desktop apps such as VS Code or Slack, Slack workspaces, exploratory QA, deriving an API client from recorded traffic, Vercel Sandbox and AWS Bedrock AgentCore browsers.

The CLI itself covers sessions, a vault for saved logins, persistent profiles, state encryption, network inspection, accessibility audits, React and Web Vitals checks, and video recording. An observability dashboard on port 4848 shows running sessions, and an MCP server mode exists for clients that prefer tools over shell commands.

## When to use it

- You want your agent to log in to a web app and test a flow end to end.
- You need data pulled from a page that only renders with JavaScript.
- You want an agent to click through an Electron app such as Slack or VS Code.
- You want browser instructions that stay in step with the installed tool version.

## When to pick something else

- Writing Playwright test files for your repo: Microsoft's Playwright CLI skill generates Playwright tests.

## What it needs

- The agent-browser CLI (npm, Homebrew or Cargo)
- Chrome, downloaded by agent-browser install or detected if already present

## Which agents it works in

Vercel documents it for Claude Code, Codex, Cursor, Gemini CLI, GitHub Copilot, Goose, OpenCode, Windsurf. The open skills CLI also installs it into 78 agents (listed with the CLI on 2026-10-01).

The SKILL.md is a stub: it tells the agent to run agent-browser skills get core and load the real instructions from the installed CLI, so the agent needs shell access and the CLI on its path.

## License

Apache-2.0.

## Related

### Related skills for browser automation and web research

- [Firecrawl](https://cookiesforai.app/skills/firecrawl/firecrawl): firecrawl is Firecrawl's agent skill for live web work through its CLI: the agent searches the web, scrapes pages to clean markdown, maps and crawls sites, and interacts with pages that need clicks or logins.