TL;DR
The ScrapeGraphAI MCP server connects any Model Context Protocol client (Claude, Cursor, and friends) to live web scraping, no code required.
- What it is: a production-ready MCP server that bridges LLMs to the ScrapeGraphAI API.
- 17 tools: scrape, extract, search, crawl, schema generation, scheduled monitors, history, and credits.
- Hosted setup: point your client at
https://mcp.scrapegraphai.com/mcpand paste your API key. - Local setup: run it as a Python package over stdio when you want full control.
What the MCP server actually does
The Model Context Protocol is a standard way for AI assistants to reach external tools and data. The ScrapeGraphAI MCP server speaks that protocol on one side and our scraping API on the other.
The practical result: you can ask Claude or Cursor to "find the pricing tiers on these five competitor sites and put them in a table," and the assistant will call the right scraping tools, wait for the results, and reason over real, current web data. No glue code, no copy-pasting HTML into a chat window.
The toolset
The server exposes 17 tools, grouped by what they do:
scrape: retrieve page content as markdown, HTML, screenshots, links, images, or a summary.extract: AI-powered structured data extraction from a URL.search: web search with structured result extraction.crawl_start/crawl_get_status/crawl_stop/crawl_resume: manage multi-page asynchronous crawls.schema: generate a JSON Schema from a plain-language prompt.creditsandhistory: check your balance and browse past requests.monitor_create/list/get/pause/resume/delete/activity: set up and poll scheduled extraction jobs.
Your assistant picks the right tool for the request. You describe the outcome; it orchestrates the calls.
Setup, the easy way: the hosted endpoint
You do not need to install anything to get started. The hosted server lives at https://mcp.scrapegraphai.com/mcp. Grab an API key from the dashboard and add the server to your client.
Cursor (remote-first):
{
"mcpServers": {
"scrapegraph-mcp": {
"url": "https://mcp.scrapegraphai.com/mcp",
"headers": {
"X-API-Key": "YOUR_API_KEY"
}
}
}
}Claude Desktop (remote over mcp-remote):
{
"mcpServers": {
"scrapegraph-mcp": {
"command": "npx",
"args": [
"mcp-remote@0.1.25",
"https://mcp.scrapegraphai.com/mcp",
"--header",
"X-API-Key:YOUR_API_KEY"
]
}
}
}Restart your client and the ScrapeGraphAI tools show up. That is the whole setup.
Setup, the local way: stdio
When you want the server running on your own machine (for air-gapped environments, custom networking, or just preference), install it as a Python package and run it over stdio:
{
"mcpServers": {
"scrapegraph-mcp-local": {
"command": "python3",
"args": ["-m", "scrapegraph_mcp.server"],
"env": { "SGAI_API_KEY": "YOUR_API_KEY" }
}
}
}Here the key travels through the SGAI_API_KEY environment variable instead of an HTTP header. Functionally it exposes the same tools as the hosted endpoint.
A first prompt to try
Once the server is connected, skip the menus and just ask:
"Use scrapegraph to scrape https://news.ycombinator.com and give me the top 10 story titles with their points and comment counts as a markdown table."
The assistant calls scrape or extract, ScrapeGraphAI does the work, and you get a clean table back in the chat. Swap in any site and any structure you need.
Where this shines
The MCP server turns your assistant into a research analyst. A few patterns we see constantly:
- Competitive intelligence: monitor competitor pages and summarize what changed.
- Lead generation: pull contact and company details from a list of sites.
- Content aggregation: gather and structure articles across many sources.
- Dataset building: crawl a section of a site and extract it into rows you can export.
Because monitors are part of the toolset, you can even ask the assistant to set up a recurring extraction and check back on it later.
Wrapping up
If you already live inside Claude or Cursor, the MCP server is the shortest path to giving those assistants real web access. Start with the hosted endpoint, paste your key, and ask for data in plain English. Move to the local install when you need the control. Either way, your AI stops guessing about the web and starts reading it.
Related Articles
- ScrapeGraphAI CLI: Web Scraping From Your Terminal - The same engine as commands you can pipe in a shell.
- ScrapeGraphAI + Vercel AI SDK: Web Tools for Agents - Give your own agents the tools your chat client gets.
- CrewAI + ScrapeGraph: Build Data Collection Agents - Multi-agent crews powered by the same engine.