Add one URL and Claude, Codex or Cursor can read any web page, search Google and pull data from Instagram, TikTok and Amazon. Clean results, sized for its context, with no browser or proxies on your side.
Find well-rated coffee shops in Austin, then tell me what recent reviews of Epoch Coffee say.
googleSearch · 10 credits
ScrapingBot googleReviews · 10 credits
Highest rated: Flora Coffee & Culture (4.9, 410 reviews), Neon Belly Coffee & Bakery (4.8) and Merfolk’s Specialty Coffee (4.7).
Epoch Coffee (4.5, 2,500 reviews): this week’s reviewers praise the espresso drinks and service, the indoor and outdoor seating and the dog-friendly patio, and one congratulates it on 20 years.
What are the hottest new books on Amazon right now?
amazonRankings · 10 credits
Amazon’s top new releases in Books:
1. A Court of Splintered Harmony (The Valkyrie Cycle, Movement I)
2. The Steps: A Memoir, 4.8 stars
3. Threshing Day (The Empyrean), 4.8 stars
What does “A Light in the Attic” cost on books.toscrape.com, and is it in stock?
extractStructuredData · 6 credits
It’s £51.77 and in stock.
Pick your assistant and paste one command or config. Then just ask: it decides when to use ScrapingBot.
Create a free account and copy your API key. You start with 100 credits, no card needed.
Paste the setup for your assistant and put your key where it says YOUR_API_KEY.
Ask a question that needs the web. Your assistant calls the right tool and answers from the result.
Run this once in your terminal. Claude Code connects right away; type /mcp in a session to see the 23 tools.
claude mcp add --transport http scrapingbot https://scrapingbot.io/api/mcp \ --header "x-api-key: YOUR_API_KEY"
Add --scope user to use it in every project, not just this one.
You don’t name tools or write requests. Your assistant reads what each tool does and costs, picks the right ones and chains them when a question needs more than one.
“Summarize https://example.com/pricing into a table of plans and prices.”
“Pull every product name, price and rating from this category page as JSON.”
“What are people saying in the comments on this TikTok video? Group them by theme.”
“Find dentists in Denver rated 4.7 or higher and tell me what the newest reviews complain about.”
“Compare these three Amazon products by price, rating and key features.”
“Which of @nasa’s recent Instagram posts got the most comments, and what were they about?”
The same APIs and the same credits you’d use from code. Credits per call are shown on the right.
A page as clean markdown (or HTML), with a screenshot if you want one.
The fields you ask for, as JSON, pulled out by AI.
Click, type, scroll and wait in a real browser, then read the page.
Web, images, videos, news, shopping, places and maps, in any country and language.
Google Maps reviews for a place, newest or most relevant first.
Profiles with follower counts, bio and links.
Posts, reels, stories, comments and replies.
Accounts, hashtags, places and posts.
Followers and following, searchable by name.
Profiles and their latest videos.
A video’s stats, caption, sound and download links.
Creators and videos by keyword.
Comments on a video, and replies.
Followers and following.
Products for a keyword with price, rating and Prime.
Everything on a product page.
Up to 20 products in one call.
Best sellers and new releases by department.
Amazon’s search suggestions, for keyword research.
Plus listCapabilities, getScrapeJob and pollJobUntilDone (free), and providerRequest for the few endpoints the tools above don’t cover. Full parameters are in the docs.
Not raw HTML: most pages are several times smaller as markdown, so your assistant reads more and spends fewer tokens. Links stay clickable.
Every result stays under about 15,000 tokens. Long lists are trimmed with a note saying what’s left and how to get it.
Ask for a screenshot and it arrives as an image, so a vision model can look at the page.
The tools only fetch public data. Private and internal addresses are blocked, and nothing is posted anywhere.
Each tool says what it costs before it runs. Calls that fail are free, and your assistant can’t spend more than your balance.
It’s a hosted server: no package, no browser, no proxies and no CAPTCHAs on your machine.
The Model Context Protocol is an open standard that lets AI assistants use outside tools. Claude, Codex, Cursor and VS Code all support it. Add ScrapingBot once and your assistant can fetch web pages and data from Google, Instagram, TikTok and Amazon whenever a question needs it.
Claude Code, the Claude app (web and desktop, as a custom connector), Codex (app and CLI), Cursor, VS Code and any other client that supports remote MCP servers over HTTP. The setup tabs above show each one.
No. The server is hosted at https://scrapingbot.io/api/mcp. You add the URL and your API key to your assistant, and that’s it.
With the same credits as the ScrapingBot API: 1 credit for a TikTok lookup or a simple page, 5 for an Instagram call or a JavaScript page, 10 for Google or Amazon. Each tool’s description includes its cost, so your assistant knows before it calls. New accounts get 100 free credits.
No. It can only spend the credits on your account. When they run out, calls stop with a clear message until you top up or upgrade. Failed calls are refunded automatically.
Your key stays in your own assistant’s settings and is sent only to scrapingbot.io over HTTPS. Treat it like a password: if it leaks, create a new key and delete the old one in your dashboard. Use the header setup where your client supports it; the URL form is for clients that only take a URL.
Results are trimmed to fit your assistant’s context, with a note saying what was left out. Ask for fewer items, the next page or a narrower question. The HTTP API always returns the full response if you need everything.
Yes. One key and one credit balance cover the MCP server and the REST API, so you can prototype in your assistant and then copy the request into your app.
100 free credits, no credit card. That’s 100 TikTok lookups, 20 Instagram calls or 10 Google searches to start.