mcp-screenshot-website-fast
An MCP server that captures web pages, tiles long ones into 1072 by 1072 chunks for Claude Vision and shoots single elements.
- 111 ★ on the repository
- Last commit
- MIT
Extended access
Why this level
- The server opens any given address in a headless browser and saves files into the chosen folder.
- The `jsEvaluate` parameter runs JavaScript on the loaded page.
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add mcp/mcp-screenshot-website-fastDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- The server opens any given address in a headless browser and saves files into the chosen folder.
- The `jsEvaluate` parameter runs JavaScript on the loaded page.
This is third-party code. Review the repository files before installing.
What it does
mcp-screenshot-website-fast captures web pages in a headless Puppeteer browser and fits the images to vision-model processing. Resolution is capped at 1072 by 1072 pixels, about 1.15 megapixels, and long pages are automatically tiled into chunks of that size. The server has three tools. `take_screenshot` captures the full page or the first screen and takes width, height, the `waitUntil` strategy and a `directory` for saving PNG files instead of base64. `capture_selector` captures an element by CSS selector. `take_screencast` takes a series of shots over a set time and can save an animated WebP. There is no cache, so shots are always fresh. The browser closes after 60 seconds of idleness and pages are closed after each shot to avoid leaks.
Who it is for. Interface developers who ask Claude Code or Cursor to compare layout with a screenshot and look at the page with their own eyes.
Good fit when
- The agent must see how a page looks after a CSS change.
- You need to capture a specific block, such as a login form, by the `#main` selector.
- You need to save all tiles of a long page to a folder for comparison.
- To check an animation you need a series of frames over ten seconds.
Not a fit when
- You need to act on a page: click, type and navigate a site.
- The page is behind authentication and the server has no sign-in.
- You need a screenshot of an application rather than a web page.
Example request
Capture the first screen of https://example.com at width 1072 and save PNG files into the shots folder, then capture the #main block separately.MCP
- Transport
- stdio
- Authentication
- not required
Limitations
Node.js 20 or newer is required, and Puppeteer downloads Chrome or Chromium itself. The maximum tile is 1072 by 1072 pixels, and wider or taller shots are not possible. The tools accept no login data, so pages behind authentication cannot be captured. A screencast captures only the top 1072 by 1072 tile. The `jsEvaluate` parameter runs JavaScript on the page, so avoid it for unverified addresses. Shots are always fetched fresh, with no cache.
How to disable. Remove the server from the client MCP configuration: for Claude Code use `claude mcp remove screenshot-website-fast`. A folder of shots, if you set `directory`, stays on disk.
FAQ
How do I connect the server to Claude Code?
With `claude mcp add screenshot-website-fast -s user -- npx -y @just-every/mcp-screenshot-website-fast`. VS Code, Cursor and JetBrains have their own commands and JSON in the README.
Why is the size capped at 1072 pixels?
That is what Claude Vision handles best: 1072 by 1072 gives about 1.15 megapixels. Long pages are cut into tiles of that size.
How do I capture one element?
Use `capture_selector` with `url` and `selector`. The wait for the element is set by `selectorTimeoutMS`, 5000 ms by default.
Where are shots saved?
Without `directory` images come back as base64. With it the tool saves PNG files with timestamps and returns file paths.
Can I record an animation?
`take_screencast` takes a series of shots with a set duration and interval, and with `directory` saves an animated WebP.
Discussion
No questions or reviews yet.
Related
MCP serversBrowserless.io MCP server: a cloud browser for the agent with page reads, site crawls, Lighthouse audits and click-driven sessions
An MCP server that lets a model drive your desktop: screenshots, mouse and keyboard through a single computer tool
An MCP server that opens pages in headless Chromium through Playwright and hands the agent the main text as Markdown, even from JavaScript-heavy sites
Hyperbrowser MCP
Hyperbrowser MCP Server
The official MCP server for Hyperbrowser cloud browsers: page scraping and crawling, structured data, Bing search and three browser agents
MCP server that reads sites on your machine and hands the agent short Markdown with links, keeping raw HTML out of context.
MCP server with DuckDuckGo search and page text extraction: several URLs can be passed in one call
README in short
The project targets command-line agents and runs through Puppeteer. The server is tuned for speed: minimal dependencies, automatic browser cleanup and a Windows-compatible launcher.