extract_structured_content
Extracts structured content from a web page using CSS selectors
LABSCO SUMMARY
You run the browser, so the rendering is yours to control.
Because the Worker is deployed to your own Cloudflare account, page loads are billed and rate-limited by you rather than by a third-party scraping service, and a page that needs JavaScript to render actually renders. The cost is the setup: nothing works until the Worker is deployed with both bindings in place, and the two most common failures — no browser binding, no SCREENSHOTS KV namespace — are both fixed in the Cloudflare dashboard, not here. CSS-selector extraction is the tool worth reaching for first; it returns a fraction of the tokens a full page fetch does.
What it is
An MCP server that talks to a Cloudflare Worker you deploy, using Cloudflare's Browser Rendering API. Pages are loaded in a real browser on Cloudflare's edge and come back processed for use as model context, which means JavaScript-rendered sites work and the traffic originates from your own Cloudflare account rather than from your machine.
What you get
- A page fetched and processed into context, with an optional cap on how much content comes back —
fetch_page - Named parts of a page extracted with CSS selectors, so a table or a heading arrives as a value rather than as prose to be parsed —
extract_structured_content - A page condensed to a length you specify, when the whole thing would not be worth the context —
summarize_content - A screenshot at a viewport size you choose — 1280 by 800 by default — either the visible area or the full page —
take_screenshot - Cloudflare's own documentation searched, with a cap on results —
search_documentation
Requirements
A Cloudflare account with Browser Rendering API access, and the included puppeteer-worker.js deployed to Cloudflare Workers with npx wrangler deploy. That Worker needs two bindings: a Browser Rendering binding named browser, and a KV namespace binding named SCREENSHOTS — screenshots fail without the second. Put the deployed Worker's URL in BROWSER_RENDERING_API. Node.js v18 or higher; the npm package name is cloudflare-browser-rendering (1.0.0), built from a clone with npm install and npm run build, or installed through Smithery.
Setup effort
Build from source — git clone https://github.com/amotivv/cloudflare-browser-rendering-mcp.git
USE CASES
01
Reading a page that a plain fetch returns empty
Load it through the Worker's real browser so client-side rendering completes, and get usable text back rather than an empty shell.
02
Pulling one number off a page repeatedly
Name the CSS selector for the element and get just that value, instead of fetching the whole page and hoping the model finds it.
03
Getting the gist of a long article
Summarise it to a set length before it enters the conversation, so a reference costs a paragraph rather than a page.
04
Showing what a page actually looks like
Capture the full page at a viewport width you choose, for a visual check that text extraction cannot give you.
05
Answering a Cloudflare configuration question
Search Cloudflare's documentation from the same session and read the relevant page through the same renderer.
5 tools
Counted once with only startup setup. Other versions, modes or optional credentials may expose a different list.
fetch_page
Fetches and processes a web page for LLM context
search_documentation
Searches Cloudflare documentation and returns relevant content
summarize_content
Summarizes web content for more concise LLM context
take_screenshot
Takes a screenshot of a web page and returns it as an image
BEFORE YOU INSTALL
What to know first
Tools you get
5 tools when we connected.
Requires
- npx on PATH
- Node.js v18 or later
- A Cloudflare account with Browser Rendering API access
- BROWSER_RENDERING_API in your environment
CONTEXT COST
Light Standing cost
Rated by the size of the 5 tool definitions that ride along on every turn, not by how many there are
INSTALLATION
Connect your client
You clone and build this one yourself.
Server setup
TERMINAL
git clone https://github.com/amotivv/cloudflare-browser-rendering-mcp.git
cd cloudflare-browser-rendering-mcp
npm install
npm run buildBefore deploying, open wrangler.toml: it already declares the browser binding and the SCREENSHOTS KV binding, but the KV namespace id it ships is the author's. Set that id (under kv_namespaces) to the id of a KV namespace in your own Cloudflare account
TERMINAL
npx wrangler deployIn the Cloudflare dashboard, give the deployed Worker a Browser Rendering binding named browser and a KV namespace binding named SCREENSHOTS
In the client block, <YOUR_WORKER_URL> is the deployed Worker's URL and <YOUR_PATH_TO_CLOUDFLARE_BROWSER_RENDERING_MCP> the full path of your clone (npm run build creates dist/index.js there)
MCP Clients
Any MCP client
ANY MCP CLIENT · JSON
{
"mcpServers": {
"cloudflare-browser-rendering": {
"command": "node",
"args": [
"<YOUR_PATH_TO_CLOUDFLARE_BROWSER_RENDERING_MCP>/dist/index.js"
],
"env": {
"BROWSER_RENDERING_API": "https://<YOUR_WORKER_URL>.workers.dev"
}
}
}
}
