What is ScreenshotMCP?
ScreenshotMCP is a Node.js MCP server that uses a headless browser to capture website screenshots. It has three tools for full-page capture, single-element capture by CSS selector and listing device presets, with wait conditions for pages that load slowly.
ScreenshotMCP is a Model Context Protocol server for capturing website screenshots with full-page, element and device-size options. You run it from a clone of the repository with Node.js, and your client launches it as a local process.
The README lists Puppeteer-installed Chrome as the browser and a rate limit plus URL validation as guards. It also describes sandboxed browser execution, so the server only accepts HTTP and HTTPS addresses.
What changes for your agent
- The agent can look at a rendered page, not only its HTML, before it changes a layout.
- Preset sizes for mobile, tablet and desktop make responsive checks a one-line request.
- An element capture takes a CSS selector, so you can screenshot one navigation bar or one pricing card.
- Wait conditions cover a selector, a custom function, a timeout or network idle.
Device presets
| Preset | Size | Scale | Mobile and touch |
|---|---|---|---|
| mobile | 375 by 667 | 2x | Yes |
| tablet | 768 by 1024 | 2x | Yes |
| desktop | 1920 by 1080 | 1x | No |
Who ScreenshotMCP is for
Best for
- Front-end work where the agent needs to compare a live page against a design at several widths.
- Quick visual checks next to Chrome DevTools MCP when you only need an image.
- Review loops that feed a capture to Design Critique MCP for scoring.
Not for
- Pages behind a login. The README lists no sign-in or cookie support.
- Hosted use. You clone it and run it yourself, and it needs Node.js 18 or newer.
- Pixel-diff testing, since it only captures images and does not compare them.
Use it with UiChemy
After you build a WordPress page with UiChemy, a screenshot server helps you check the result at phone and desktop widths. Figma to WordPress converts designs, and Composer edits the page. Click Get Started on uichemy.com.
Connect ScreenshotMCP: Full-Page and Element Website Screenshots for Agents to your AI client
git clone https://github.com/upnorthmedia/ScreenshotMCP && cd ScreenshotMCP && npm installGet the code
Clone or download the repository and move into its folder. The README does not list a package install command.
Prepare the environment
Rename
.env.exampleto.env, then runnpm install. Chrome or Chromium comes with Puppeteer.Start the server
Run
npm startto try it. Your MCP client normally starts the server itself from the config entry.Add the config entry
Paste the JSON above into your client's MCP settings and point
argsat your copy ofindex.js. The README also shows aclaude mcp addline for Claude Code.Ask for a capture
Try a request such as a mobile screenshot of your site, or a screenshot at 1024 by 768.
Client configuration
{
"mcpServers": {
"screenshot-full-page-mcp": {
"command": "node",
"args": ["/path/to/screenshot-full-page-mcp/index.js"]
}
}
}Endpoints and ports
index.jsLaunched with node.envBrowser timeout, concurrency, rate limitsNode.js 18 or newerAt least 2GB RAM and 500MB diskWhat you can do with it
- Full-page captureThe page is scrolled automatically so the whole length is captured.
- Element captureTarget a single section by CSS selector and skip the rest of the page.
- Custom viewportSet width and height from 100 to 5000 pixels, plus scale factor, mobile and touch emulation.
- Wait conditionsWait for a selector, a function, a timeout or network idle before the shot.
- Concurrency limitAn environment setting caps simultaneous captures to protect memory.
- Input checksURLs must be HTTP or HTTPS, selectors are sanitized and requests are rate limited.
- Debug modeSet headless to false in the environment file to watch the browser while troubleshooting.
Tools it exposes
capture_screenshot
Captures a full-page screenshot, with viewport, wait conditions, delay and navigation options.
capture_element
Captures one element chosen by a CSS selector.
list_device_presets
Lists the device presets and their settings.
Frequently asked questions
What is ScreenshotMCP?
It is an open-source MCP server that captures website screenshots. An AI agent can request a full page, one element or a device-sized view, and the server returns the image using a headless browser launched through Puppeteer.
How do I install ScreenshotMCP?
Clone the repository, rename the example environment file to .env, run npm install, then add a node command pointing at index.js to your MCP client config. The README lists no registry package, so installation is from source.
Which device sizes does it support?
Three presets ship with it: mobile at 375 by 667, tablet at 768 by 1024 and desktop at 1920 by 1080. You can also set a custom width and height between 100 and 5000 pixels.
Can it screenshot just one element?
Yes. The capture_element tool takes a URL and a CSS selector and captures only that element, using the same viewport options as the full-page tool, so you can check one section at phone width.
Can it wait for a page to finish loading?
Yes. You can wait for a selector, a function, a fixed timeout or network idle, add a further delay, and choose the navigation completion condition before the screenshot is taken.
What are the system requirements?
The README lists Node.js 18 or newer, Chrome or Chromium installed automatically with Puppeteer, at least 2GB of memory and 500MB of disk space. A concurrency setting limits how many captures run at once.
How is it different from a browser DevTools server?
ScreenshotMCP does one job, capturing images. A fuller browser tool such as Chrome DevTools MCP inspects and debugs pages, so many teams use a small capture server only when an image is enough.


