Skill Finder
The best Claude skills for browser automation
Start with browser-use from Browser Use: at 86 it has the top score here without a high-risk scan warning, and it lets Claude control your own Chrome or a paid cloud browser. agent-browser scores 95 with over 1.2 million installs, but at least one scanner rated it high risk. The list also covers Firecrawl, Tavily, OpenCLI, agent-device for mobile apps and webapp-testing.
The 9 skills made for this job
-
1
Get agent-browseragent-browser
For developers who want Claude to drive Chrome from the terminal to click, fill forms and take screenshots. Skip it if you cannot install a global npm tool.
-
2
Get browser-usebrowser-use
For developers who want Claude to drive their own Chrome, or a hosted cloud browser, through the DevTools protocol. Skip it if a plain fetch can read the page.
-
3
Get firecrawl-searchfirecrawl-search
For people who want Claude to search the web from the terminal and keep results with excerpts as JSON. Skip it if you do not want to spend credits per search.
-
4 Get firecrawl-agent
firecrawl-agent
For people who need structured data pulled from sites that take many pages to navigate. Skip it if one page scrape is enough or you won't pay for Firecrawl credits.
-
5 Get firecrawl-scrape
firecrawl-scrape
For people who have a URL and want its content as markdown or typed JSON. Skip it if you still need to find the pages or must click through forms.
-
6
Get agent-deviceagent-device
For mobile and TV app developers who want Claude to tap through apps on simulators and devices. Skip it if you only test web pages in a browser or cannot install the agent-device CLI.
-
7
Get tavily-searchtavily-search
For anyone who wants Claude to check current facts, news or sources on the web through Tavily. Skip it if you have no Tavily access or your work never needs live web data.
-
8
Get opencli-usageopencli-usage
For developers who use OpenCLI and want Claude to find and run its site adapters and Chrome tools. Skip it if you don't have OpenCLI set up.
-
9 Get opencli-browser
opencli-browser
For developers who want Claude to work inside their logged-in Chrome through OpenCLI and capture site APIs. Skip it if you'd rather keep agents out of your real browser profile.
Also useful for browser
Their main job is something else, but they help with this one.
-
Get webapp-testingwebapp-testing
For developers who want Claude to click through a web app running on their own machine and see what breaks. Skip it if the site is only live on a remote server.
-
Get next-dev-loopnext-dev-loop
For Next.js developers on Turbopack who want each edit confirmed in a real browser. Skip it if you're below Next.js 16.3, on webpack, or not running next dev.
-
Get create-payment-credentialcreate-payment-credential
For people who want Claude to buy things for them from a Link wallet, approving each spend in the Link app. Skip it if you would not let an agent handle card numbers or checkout.
How to choose
browser-use sends short Python snippets to a background daemon that controls Chrome with remote debugging turned on. It stops to ask at login walls and can start a paid Browser Use Cloud browser for parallel or bot-sensitive jobs. It needs the browser-use CLI, installed with uv.
agent-browser is a short stub that loads a usage guide matching the installed CLI and picks elements by refs such as @e1. It needs Node.js and the agent-browser CLI. At least one scanner rated it high risk, so read its scan results before you install it.
If you only need page content, skip the clicking. firecrawl-search saves search results as JSON, firecrawl-scrape turns known URLs into markdown or schema-shaped JSON, and firecrawl-agent runs a hosted job across many pages; all three spend Firecrawl credits. tavily-search picks search, extract, map or crawl and needs a Tavily account.
The rest fit narrower setups. opencli-browser and opencli-usage drive Chrome and website commands through OpenCLI; at least one scanner rated opencli-usage high risk. agent-device taps through iOS, Android and TV apps on simulators and devices.
Secondary picks come from other jobs. webapp-testing, from Anthropic, writes Python Playwright scripts that click through a web app running on your own machine. next-dev-loop checks Next.js edits in a running app, and create-payment-credential handles Link checkout.