MCP server

Give Your AI Agent Web Access

rasterly ships as an MCP server, live in the official registry — npx -y rasterly-mcp and Claude, Cursor or any MCP client can read, see, extract from and crawl the live web.

An LLM agent's built-in fetch gets raw HTML — often an empty <div id="root"> for JS-rendered sites, and nav/ad noise for everyone else. rasterly-mcp runs a real headless browser for every call, so the agent gets what's actually on the page: clean Markdown, a real screenshot, structured data, or a print-ready PDF.

Add it to your MCP client

Works in Claude Desktop, Claude Code, Cursor, Windsurf, and any other MCP client. Add this to your config (claude_desktop_config.json, .mcp.json, or the client's MCP settings) and restart:

mcp config
{
  "mcpServers": {
    "rasterly": {
      "command": "npx",
      "args": ["-y", "rasterly-mcp"],
      "env": { "RASTERLY_API_KEY": "sk_live_your_key" }
    }
  }
}

Get a free key at rasterly.dev first — 100 renders/month, no card required.

Tools

ToolWhat the agent gets
read_urlClean, LLM-ready Markdown — JS rendered, nav/ads stripped.
capture_urlA screenshot the agent can actually see (image content, full_page optional).
url_to_pdfA print-ready A4 PDF, returned as a base64 resource.
extract_urlStructured JSON — JSON-LD, OpenGraph, tables, images, links.
crawl_urlA whole site as Markdown, following same-origin links (great for RAG).

Try it

Once it's connected, just ask your agent:

Why rasterly

The MCP server is a thin wrapper over the same rasterly API — one browser process reused across requests, so a warm render is about a second of CPU. That keeps calls from $0.90 per 1,000 with a 100/mo free tier, so an agent that reads or screenshots a dozen pages in a session doesn't run up a real bill.

FAQ

How do I add rasterly to Claude Desktop, Claude Code or Cursor?
Add the rasterly entry to your MCP config (claude_desktop_config.json, .mcp.json, or your client's MCP settings) with command npx, args ["-y", "rasterly-mcp"], and your RASTERLY_API_KEY in env. Restart the client and the tools appear.
What tools does the MCP server expose?
Five: read_url (clean Markdown), capture_url (screenshot image), url_to_pdf (PDF resource), extract_url (structured JSON), and crawl_url (multi-page Markdown for RAG).
Is this different from an agent just calling fetch()?
Yes — fetch() gets raw HTML, which is often empty for JS-rendered sites and full of nav/ad noise for everyone else. Each rasterly tool runs a real browser first, so the agent gets the content that's actually on the page.
Is rasterly-mcp in the official MCP registry?
Yes — it's published to the official Model Context Protocol registry as io.github.jhonychain/rasterly-mcp, and to npm as rasterly-mcp, so npx -y rasterly-mcp always pulls the latest version.
Get a free API key Try the live demo
More: Best screenshot APIs URL to Markdown API HTML to PDF API Web scraping / structured data API Website crawler for RAG