browserbase/skills: Give Claude Code a Full Browser Automation Toolkit

An AI agent that cannot drive a browser hits a wall the moment a task leaves the filesystem: login-gated sites, JavaScript-heavy pages, your localhost dev server. browserbase/skills is the official Browserbase collection that gives Claude Code a whole browser automation toolkit — not one skill, but seventeen, spanning raw browsing, trace capture, UI testing, and research report generation.

Why This Skill Matters

Most browser skills give you one thing: open a page, read the DOM. This collection covers the full lifecycle around that. The browser skill drives remote Browserbase sessions with Browserbase Identity, Verified browsers, CAPTCHA solving, and residential proxies. browser-trace captures a full DevTools-protocol stream and bisects it into per-page searchable buckets. ui-test does adversarial testing — it analyzes your git diffs to decide what to test, or explores the whole app to find bugs. There are also research-oriented skills: company-research deep-researches ICP-matched companies into a report, and competitor-analysis compiles a browsable HTML report with a feature/pricing matrix.

The repo sits at 3.7k stars and ships as a Claude Code plugin marketplace, so installation and updates follow the standard plugin flow.

Installation

The generic installer handles most agents:

npx skills add browserbase/skills

On Claude Code, the plugin path is the recommended one. Add the marketplace:

/plugin marketplace add browserbase/skills

Then install the plugin:

/plugin install browse@browserbase

If you prefer the manual interface: type /plugin, select option 3. Add marketplace, enter browserbase/skills as the source, pick the browse plugin, and hit Install now. Restart Claude Code afterwards for the changes to take effect.

Real Workflow: QA Your Local Dev Server

The README's own example is the one you will use most. You have an app running on localhost and want the agent to find and fix bugs.

1. Install the plugin and restart Claude Code.
2. Start your dev server as usual, then prompt:

QA test http://localhost:3000 and fix any bugs you encounter

3. For local and localhost work, pass --local on the first browser command (for example, browse open http://localhost:3000 --local) so the agent starts a clean, isolated browser instead of reaching for a remote session.

The agent opens the page, clicks through flows, reports what breaks, and can apply fixes since it is already in your repo.

Real Workflow: Summarize a Discussion Thread

For read-and-summarize tasks on public sites, just describe the goal:

Go to Hacker News, get the top post comments, and summarize them

If the agent should reuse your existing local Chrome session — cookies and login state included — use --auto-connect on the browser command instead. For sites that need your authenticated profile on remote sessions, the cookie-sync skill syncs cookies from local Chrome into a Browserbase persistent context.

Tips

• Use --local for localhost and --auto-connect when the agent should reuse your existing Chrome session.
• If Chrome is not found, install it for your platform — on Linux: sudo apt install google-chrome-stable.
• To refresh cookies from your main Chrome profile, remove the .chrome-profile directory and retry.
• safe-browser builds agents whose only browser tool is CDP-gated with domain allowlist enforcement — use it when scope discipline matters more than flexibility.
• browser-to-api can turn a site's observed HTTP traffic into a best-effort OpenAPI 3.1 spec from a browser-trace capture.

When Not to Use This

If the target page is static HTML or JSON, the bundled fetch and search skills — or plain curl — are faster than launching a browser session. And if you are running large-scale parallel scraping, talking to the Browserbase API or Stagehand directly will beat agent-mediated clicks on cost and determinism.

See the leaderboard for more skills.