Web Scraping MCP servers

Browsers, crawlers and page extraction.

Web scraping and browser MCP servers give an assistant a way onto the open web. That covers two quite different jobs: fetching a page and reading it, and driving a real browser that runs JavaScript, holds a session and can click things.

Fetch-style servers are the cheap option. They retrieve a URL, strip it to readable text or Markdown, and hand it back. For documentation, articles and anything server-rendered, that is all you need and it costs one HTTP request.

Browser-automation servers are the expensive, capable option. They start a real browser engine, so single-page apps render, logins persist and forms submit. That is what you need for testing a flow, screenshotting a page or extracting data that only exists after JavaScript has run.

Web Scraping MCP servers in this directory

What people build with browser and scraping MCP servers

Reading documentation the model has never seen

A fetch server lets an agent read the current docs for a library released after its training cut-off, instead of confidently describing an API that no longer exists.

End-to-end testing described in English

Describe the flow — sign in, add to basket, check out — and let the browser server drive it, reporting where it broke and what the console said.

Extracting structured data from a rendered page

Pull a table, a price list or a set of listings out of a page that only assembles itself client-side.

Visual checks and screenshots

Capture a page at several viewport widths and have the agent describe what actually shifted, rather than reasoning about the CSS in the abstract.

Choosing a browser or scraping MCP server

  • Use a fetch server if the content is in the HTML. It is faster, uses far less memory, and does not need a browser binary installed.
  • Use a browser-automation server if the content appears after JavaScript, or if the task involves interaction rather than reading.
  • Check what the server downloads on first run. Browser engines are large, and the first install can pull hundreds of megabytes before anything works.
  • Prefer servers that return Markdown or plain text over raw HTML. Raw markup burns context on tags the model has no use for.

Before you connect one

Respect the sites you point these at: check robots.txt and terms of service, rate-limit your own requests, and do not use a browser server to work around a paywall or an authentication wall you do not have the right to pass. A page fetched into a conversation is also a page whose contents you have handed to a model — treat untrusted page text as untrusted input, because a page can contain instructions aimed at your agent.

Frequently asked questions

What is the difference between a fetch server and a browser server?

A fetch server makes an HTTP request and converts the response to text — fast, light, and blind to anything rendered client-side. A browser server runs an actual browser engine, so JavaScript executes, sessions persist and the agent can click, type and screenshot.

Can these log into a site for me?

Browser-automation servers can drive a login form, and some persist the session between calls. Whether you should is a separate question: it means credentials in your config and an agent acting as you on that site.

Do I need a headless browser installed?

For browser-automation servers, yes — most download their own engine on first run rather than using the browser you already have. Fetch servers need nothing beyond a network connection.

Is scraping with an MCP server legal?

The same rules apply as any other scraping: public data, reasonable rates and the site's own terms. An MCP server changes who is issuing the request, not what you are allowed to request.

Install any of these

Every server on this page carries a one-line install command on its own listing. Paste it into your client config and restart — the web scraping tools appear in your next session.

Or keep browsing: all MCP server categories · the full MCP server list · Agent Skills.