The proxy tier is the lever that matters: residential is the default, and stealth exists for sites with serious bot detection at a higher cost per call. Content sandboxing is worth turning on by default — it wraps whatever the page said in explicit boundaries so scraped text is not read as instructions.
A server over the WebScraping.AI API with seven tools: ask a question about a page, extract named fields, or take the HTML, text or a selector's contents.
- webscraping_ai_question answers a question about a page's content
- webscraping_ai_fields extracts structured fields you name
- webscraping_ai_html and webscraping_ai_text return the rendered HTML or plain text
- webscraping_ai_selected and webscraping_ai_selected_multiple return the contents of one or many CSS selectors
- webscraping_ai_account reports your usage
- Per-call control: js and js_timeout for rendering, wait_for to hold until a selector appears, js_script to run your own code on the page
- Proxy choice per call — datacenter, residential or stealth — plus country selection across us, gb, de, it, fr, ca, es, ru, jp, kr and in, or your own custom_proxy
- Device emulation across desktop, mobile and tablet; error_on_404 and error_on_redirect turn silent misses into errors
- Content sandboxing wraps scraped text in explicit boundaries so a page cannot pass itself off as instructions
WEBSCRAPING_AI_API_KEY is required for every operation; the free trial includes 2,000 credits with no card. Run it with npx and the npm package webscraping-ai-mcp. Optional tuning: WEBSCRAPING_AI_CONCURRENCY_LIMIT (default 5), WEBSCRAPING_AI_DEFAULT_PROXY_TYPE (default residential), WEBSCRAPING_AI_DEFAULT_TIMEOUT (default 15000, max 30000) and WEBSCRAPING_AI_ENABLE_CONTENT_SANDBOXING. A hosted remote server at https://mcp.webscraping.ai/mcp exists if you would rather not run anything locally.
One command plus a key — env WEBSCRAPING_AI_API_KEY=your_api_key npx -y webscraping-ai-mcp, then supply credentials