Give Claude access to any web page
Claude cannot open most of the web on its own - anti-bot systems block it and modern pages need a real browser. Our MCP server hands Claude a working browser and your own API key, so it can read any page, parse it, and search Google for you.
500 free requests on signup · no card needed
Why you need it
Without it, Claude is locked out of most of the web
Blocked before it reads a word
Cloudflare, DataDome, PerimeterX and Akamai turn away plain fetches with a 403, a captcha or an endless challenge loop.
Empty pages
Most modern sites render their content with JavaScript. Without a real browser, the assistant gets an empty shell with no prices, listings or reviews in it.
The wrong country
Prices, availability and search results change by location. Fetching from one place gives you one answer, not the one your users see.
What Claude gets
The right tool for each job, picked automatically
You never call them by name. Ask in plain language and Claude chooses the right one.
fetch_markdown Any page as clean Markdown: the main content with links and tables, without scripts, menus or footer. Usually 2-25% the size of the HTML, so Claude can read far more pages in one conversation.
search_and_read A Google search plus the top 1-5 result pages as Markdown in one call, with Google's AI Overview when there is one.
fetch_html The fully rendered HTML of any URL, past anti-bot protection. Optional country, wait rules and extra render time. To click, type or scroll first, browse_page runs browser steps, and list_elements finds the selectors.
fetch_parsed The same page returned as clean, structured JSON instead of raw HTML - product details, article content, listings.
google_search Google organic results as JSON, from any country, up to ten pages deep.
list_plugins / run_plugin 100+ ready-made site plugins (Amazon, Walmart, Zillow, Google Maps, YouTube and more) that return structured JSON. The 15 most-used ones are also tools of their own, so Claude calls them in one step.
Setup
Connected in about a minute
Pick where you use Claude. Your own key, your own account, your own quota.
claude.ai (web & mobile)
Settings → Connectors → Add custom connector. Paste the URL below, then open Advanced settings and set the OAuth Client ID. You sign in with your ScrapeUnblocker account - no key to copy.
https://mcp.scrapeunblocker.com/mcp OAuth Client ID: 5BM5Wk2dE4ABkDITmuKfemPLnn3QQ8jd
Claude Code
One command in your terminal. Nothing to install.
claude mcp add --transport http scrapeunblocker \ "https://mcp.scrapeunblocker.com/mcp?key=YOUR_API_KEY"
Claude Desktop
Settings → Developer → Edit Config, paste this into claude_desktop_config.json and restart. Needs Node.js 18+.
{
"mcpServers": {
"scrapeunblocker": {
"command": "npx",
"args": ["-y", "scrapeunblocker-mcp"],
"env": { "SCRAPEUNBLOCKER_KEY": "YOUR_API_KEY" }
}
}
} Replace YOUR_API_KEY with the key from your dashboard. Any other MCP client (Cursor, VS Code, n8n, LibreChat and the rest) works too - it is a standard Streamable HTTP and stdio server.
Connecting another agent, an SDK or a workflow tool? See every way to connect →
In practice
Things you can ask once it is connected
FAQ
Common questions
Is the MCP server free?
The server itself is free and open source. The requests it makes run on your ScrapeUnblocker account, and every account starts with 500 free requests - no card needed.
Does my API key go through your servers?
Running it locally, the key never leaves your machine - the server talks to our API directly. On the hosted connector you can sign in with OAuth instead, so there is no key to paste at all.
Which assistants does it work with?
Anything that speaks the Model Context Protocol: claude.ai on web and mobile, Claude Desktop, Claude Code, Cursor, VS Code, n8n, LibreChat and others.
Can Claude fill forms or click through a site?
Yes, within a page. browse_page runs browser steps (click, type, select, press a key, scroll, wait) that run in our remote browser before the page is captured, and list_elements shows the selectors to use. Nothing runs on your machine. Sign-in and payment pages are not fetched, and steps do not fill password or card fields.
How many requests does one question use?
One request per page fetched or search run. A question that asks Claude to compare five pages costs five requests, and search_and_read counts its search plus each page it reads.
Give your assistant the whole web
Create a free account, copy your key, and connect it in about a minute.