n8n is great at moving data between APIs. It is not great at scraping the open web: the HTTP Request node returns raw HTML, does not run JavaScript, and gets turned away by Cloudflare or DataDome. So the moment a workflow needs the readable content of a page, or structured fields from a site that fights back, you are stuck gluing together a headless browser somewhere else.

WebReaper now closes that gap. As of v11.2.0 it ships a Streamable HTTP MCP server, and n8n's MCP Client node speaks exactly that. Point n8n at a running WebReaper server and your workflows get five scraping tools, including JS-rendering and auto-escalating stealth, with no browser code of your own.

Run the server

The container bakes a headless Chromium, so JS-rendered pages work out of the box:

docker run -p 8080:8080 -e WEBREAPER_MCP_TOKEN=your-secret \
  ghcr.io/alex-on-ai/webreaper-mcp-http:latest

WEBREAPER_MCP_TOKEN is required: the server refuses to start on a public interface without one, so you never accidentally expose an open scraper. Keep that token; n8n sends it as a bearer credential.

If you run n8n in Docker too, drop both into one docker-compose file so they share a network. There is a ready-made example in the repo under WebReaper.Mcp.AspNetCore/docker-compose.example.yml.

Wire it into n8n

Add an MCP Client node to your workflow (or an MCP Client Tool node if you are attaching it to an AI Agent). Configure three fields:

Server Transport: HTTP Streamable
MCP Endpoint URL: your server URL. Use http://host.docker.internal:8080 when n8n runs in Docker on the same host, http://webreaper-mcp:8080 on a shared compose network, or http://localhost:8080 for a native n8n.
Authentication: Bearer, with the value of your WEBREAPER_MCP_TOKEN.

n8n fetches the tool list automatically. Pick a tool, set its inputs, and run.

To skip the clicks, import this starter workflow (a manual trigger that scrapes one URL to Markdown) and set your URL and token after import:

Download the starter workflow

The node landed in recent n8n; if your version shows it as unrecognized, add the MCP Client node by hand with the three settings above.

The five tools

scrape a page to clean, LLM-ready Markdown.
map a site to discover its URLs (sitemap plus on-page links).
extract structured fields with a CSS-selector JSON schema.
extract_with_prompt structured data by a plain-language instruction, with an LLM (set WEBREAPER_LLM_* on the server).
crawl a whole site: a bounded on-domain sweep, one Markdown record per page.

Two patterns cover most workflows. For a known set of pages, call map, fan the URLs out with a Loop node, and scrape each. For "find and read whatever is relevant," attach the MCP Client Tool node to an AI Agent and let it call the tools itself.

Why route scraping through WebReaper

Every tool runs on WebReaper's loader, so an n8n workflow inherits the parts that are painful to build yourself:

JS-rendered pages: set browser: true and the baked Chromium renders the page before extraction.
Bot protection: a plain fetch auto-climbs HTTP to a browser on a detected block (Cloudflare, DataDome, PerimeterX). For a shared browser pool, point WEBREAPER_CDP_URL at a browserless sidecar.
Clean output: Markdown and structured JSON, not raw HTML you have to parse in a Function node.

One note on crawl: it is a single long call with no progress feedback, bounded to 50 pages by default. For a large site, prefer the map-then-scrape pattern so each step stays short.

Get started

It is free and MIT-licensed. Run the container, wire the node, and your n8n workflows can read the web.