mcps
E T F P

Recipe

Let Claude search and browse the live web

Real-time search, page scraping, and full browser automation — Claude works with what's on the web right now.

for Devs and QA who want a sanity check on a flow without standing up Selenium

Hand-checked recipe · curated by Ilya Gindin

15 minutes

to run

intermediate

difficulty

4

connectors

What you'll need

This recipe runs in 3 steps

  1. 1

    Add live search

    Say to your AI

    Install a web-search MCP (Exa or Tavily). Find the three most recent posts about MCP security and summarize each in one line.

    You'll get: Three current results with one-line takeaways and links.

  2. 2

    Scrape a source

    Say to your AI

    Use Firecrawl to pull this docs page into clean markdown and extract just the configuration options.

    You'll get: A tidy markdown extract of the relevant section.

  3. 3

    Act on a page

    Say to your AI

    Use Playwright to open our staging site, run the signup flow with a test email, and screenshot each step.

    You'll get: Step-by-step screenshots and a pass/fail on whether signup completed.

You're done when

Claude stops being frozen at its training cutoff — it can find current information, read real pages, and operate a browser.

Why this workflow exists

A model alone can’t see today’s web. Search servers (Exa, Tavily, Brave) give it current results; scrapers (Firecrawl) turn messy pages into clean text; browser drivers (Playwright, Browserbase) let it actually click through a flow. Combine search to find, scrape to read, and browser to act — that covers almost every “go look at the web for me” task.

People ask their AI

“let Claude search the web”“scrape a page with Claude”“automate the browser with Claude”“what MCP for web search or scraping”

Related