Add a web-scraping API to Cursor. No glue code.
The fastest setup is a remote MCP server. Paste one JSON block — or click Add to Cursor — and the agent can call Plainscrape as a tool.
This is the Cursor-specific setup. The shared MCP reference, including the five paste-ready prompts, lives on the MCP page.
Plainscrape is not an npx process on your laptop. The server is https://www.plainscrape.com/mcp. Auth is x-api-key. Five tools, same verbs as REST: extract, extract_page, answer, get_results, account. There is no search tool and no crawl-the-whole-site tool. max_pages is clamped at 10.
Add to Cursor
Add to Cursor or paste the config
{ "mcpServers": { "plainscrape": { "url": "https://www.plainscrape.com/mcp", "headers": { "x-api-key": "ps_live_YOUR_KEY" } } } }
Restart Cursor. Ask: “Use Plainscrape extract on https://books.toscrape.com/ for every book with title, price, and rating, max_pages 1.”
What you get back
Structured JSON for the instruction — not a markdown dump of the DOM. JavaScript-rendered pages are fetched for you. Failed pages are not billed. sources_used and cost_usd are often empty today; the meter you can trust is pages_used.
FAQ
Paste a remote MCP config (or use the Add to Cursor button). Plainscrape is an HTTP server at https://www.plainscrape.com/mcp — no npx package, no local process. Your agent calls five tools: extract, extract_page, answer, get_results, account.
Structured JSON for the question you asked. The model does not have to parse HTML. There is no search tool and no site-wide crawl; max_pages is clamped at 10.
No. Plainscrape is already running over HTTPS. You add a URL and an x-api-key header. There is no local Node process to keep alive.
The same meter as curl: per successful page fetched. Failed and blocked pages are free. New accounts get a one-time 100-page trial (14 days, no card).