Model Context Protocol

ScrapingBot MCP Server

Give Claude, Cursor and other AI agents live web data. They scrape pages, products, real estate listings, travel pages, Google results and social media through ScrapingBot, right from the conversation.

Get your API key Connect to claude.ai Add to Claude Desktop View on GitHub500 free credits per month, no card required.

At a glance

Install: npx -y @sostag/scraping-bot-mcp
Clients: claude.ai, Claude Desktop, Claude Code, Cursor
Tools: 10 tools, every API
Cost: API credits, from 1 per call
Free plan: 500 credits / month
Runs: Hosted (mcp.scraping-bot.io) or on your computer
Requires: claude.ai, Claude Desktop: nothing
Source: Open source, MIT

Connect claude.ai to the hosted server with your ScrapingBot account, or run the server on your computer with your API key. Each tool call uses credits like a regular API call; failed scrapes are refunded.

Set up in two minutes

Step 1
Create your account
A free account is enough. With claude.ai you only sign in; the other clients use your username and API key from the dashboard.
Step 2
Add the server
claude.ai: add a custom connector. Claude Desktop: open the extension file. Claude Code: run the command. Other clients: add the configuration block.
Step 3
Ask your agent
Restart the client: the ScrapingBot tools appear, and the agent calls them when it needs live web data.

claude.ai (web and mobile): sign in with ScrapingBot

In Claude, open Settings → Connectors and choose Add custom connector.
Name it ScrapingBot and paste this URL: https://mcp.scraping-bot.io/mcp
Click Connect, sign in to ScrapingBot and allow Claude. No API key to copy: disconnect it at any time in your account settings.

Custom connectors depend on your Claude plan. Once added, the connector also works in the Claude mobile apps.

Claude Desktop: one-click installDownload the extension and open it: Claude Desktop installs it and asks for your username and API key. Nothing else to install.

Download scraping-bot.mcpb

claude mcp add scraping-bot \
  -e SCRAPING_BOT_USERNAME=your_username \
  -e SCRAPING_BOT_API_KEY=your_api_key \
  -- npx -y @sostag/scraping-bot-mcp

Replace your_username and your_api_key with the credentials from your dashboard, then restart the client. For other clients, use the same command, arguments and environment variables, or the hosted server's URL with the two headers.

Available tools

The agent picks the tool from your request. Page tools take a url and the optional useChrome, premiumProxy, proxyCountry and premiumFallback options.

Tool	What it returns	Credits
`scrape_raw_html`	HTML of any public page, with optional JavaScript rendering	From 1
`scrape_retail`	Product: title, price, currency, images, brand, stock, EAN	From 1
`scrape_real_estate`	Listing: price or rent, surface, rooms, location, images	From 1
`scrape_booking`	Hotel, rental or search page: Booking.com, Airbnb, Expedia, Hotels.com, Abritel	From 1
`scrape_grocery`	Grocery product: price, image, brand, nutritional values	From 1
`scrape_search_engine`	Google or Bing results as JSON: organic results, related searches, questions	30
`scrape_social_media`	Public Instagram, TikTok, Facebook, LinkedIn, X and Threads profiles, posts and searches	30
`get_social_media_result`	Result of a social media scrape that took longer than a minute	0
`get_account`	Your plan, credits used and remaining, and when they reset	0
`get_usage_history`	Credits spent per day, requests per API, success rate and latest requests	0

Example

Ask in plain language: the agent calls the tools, reads the JSON and answers.

You: Here are two product pages for the same headphones, on Amazon.fr
     and on Fnac. Which one is cheaper, and are both in stock?

Claude: I'll scrape both product pages.
  → scrape_retail  url=https://www.amazon.fr/dp/...
  → scrape_retail  url=https://www.fnac.com/a.../...

  Amazon.fr: <price> EUR, <stock>
  Fnac:      <price> EUR, <stock>
  Then the agent compares the prices and answers.

Try a scrape before connecting your agentRun a request in the Live Test and see the JSON your agent will receive.

Open the live test

Credits, failures and limits

Case	Credits	Notes
Page tools (HTML, product, real estate, travel, grocery)	1 credit	5 with useChrome, 10 with premiumProxy, 25 with both. Some sites always get the options they need. A request blocked by the site is retried with a premium proxy, billed at the premium price only if it succeeds.
Search engine	30 credits	One results page per call.
Social media	30 credits	Waiting for a pending result with get_social_media_result is free.
Failed, blocked or timed-out scrape	Refunded	The agent receives the error message and can retry with other options.

MCP server or REST API?

	MCP server	REST API
Who calls the API	Your AI agent, when it decides to	Your code
Setup	One command or configuration block	HTTP requests in your application
Best for	Research, comparisons, one-off extractions	Scheduled jobs and large volumes
Credits	Same API, same credits	Same API, same credits

Building your own integration? Read the API documentation.

Questions about the MCP server

Is this the official ScrapingBot MCP server?: Yes. It is maintained by the ScrapingBot team and its source code is public on GitHub, under the MIT license.
Do I need to install anything?: With claude.ai, nothing: add the connector and sign in. With Claude Desktop, nothing either: open the scraping-bot.mcpb extension and enter your username and API key. With other clients, Node.js 18 or later (npx downloads the server the first time it starts), or the hosted server's URL.
Where is my API key stored?: With claude.ai, nowhere: you sign in to ScrapingBot and allow the connection, and Claude gets a token instead of your key. You can disconnect it at any time in your account settings. With the other clients, in their configuration on your computer (Claude Desktop keeps it in its secure storage); the server only sends it to api.scraping-bot.io and never logs it.
How do I control what my agent spends?: Every tool call is a regular API call and uses your credits. Each tool tells the agent its cost, the agent asks before large or costly batches, and it can check your remaining credits and usage with get_account and get_usage_history. Failed scrapes are refunded.
Why do social media scrapes sometimes return a responseId?: Social media pages take from a few seconds to several minutes. The tool waits about a minute; after that the agent gets a responseId and fetches the result later with get_social_media_result, at no cost.
Which AI clients are supported?: claude.ai (web and mobile apps) through the hosted server, Claude Desktop, Claude Code, Cursor, Windsurf and the other MCP-compatible editors and agent frameworks, locally or with the hosted server's URL.
How do I pin a version?: Add the version to the package: @sostag/[email protected]. The versions are listed on npm and in the GitHub releases. The hosted server always runs the latest version.

Source code, changelog and issues: sostag/scraping-bot-mcp on GitHub.

Give your AI agent live web data

500 free credits every month. No card required, cancel anytime.

Get your API key Talk to us