The full upstream README, mirrored here for reference. Install config, tool schemas, adoption signals, and an original overview live on the Rasterly listing page.
Give any AI agent eyes and a reader for the live web. An MCP server that lets your agent turn a URL into clean Markdown, a screenshot it can actually see, or a PDF — backed by rasterly, a real headless-Chromium rendering API.
Works in any MCP client: Claude Desktop, Claude Code, Cursor, Windsurf, and others.
| Tool | What the agent gets |
|---|---|
read_url | The page's main content as clean, LLM-ready Markdown — JS is rendered, nav/ads/boilerplate stripped. Use it to read a URL. |
capture_url | A screenshot returned as an image, so the agent can see the page (layout, charts, anything visual). full_page optional. |
url_to_pdf | A print-ready PDF of the page (base64 resource). |
Why an agent wants this: a plain fetch() gets you raw HTML that a modern, JS-rendered
site won't even populate. read_url runs a real browser and hands back just the article —
far fewer tokens, far more signal.
Get a key at rasterly.dev (free tier: 100 renders/month), then add the server to your MCP client.
Claude Desktop / Claude Code / Cursor — add to your MCP config
(claude_desktop_config.json, .mcp.json, or the client's MCP settings):
Restart the client. Your agent now has read_url, capture_url, and url_to_pdf.
Ask your agent things like:
| Env var | Default | Notes |
|---|---|---|
RASTERLY_API_KEY | – | Your rasterly key (sent as X-Api-Key). |
RASTERLY_API_URL | https://rasterly-production.up.railway.app | Override to point at a self-hosted rasterly. |
MIT · built on rasterly