Screenshot Website Fast MCP Server

Overview

The @just‑every/mcp-screenshot‑website‑fast server is a purpose‑built tool for AI vision workflows that need reliable, high‑quality images of web pages. It solves the problem of generating consistent visual inputs for models such as Claude Vision by automatically limiting resolution to a 1072 × 1072 pixel square—exactly the size that maximizes compatibility and performance for 1.15‑megapixel inputs. By tiling full pages into these chunks, the server ensures that even very long or complex sites can be captured without exceeding model limits or memory constraints.

For developers, this means a single, lightweight MCP endpoint that can be invoked from any AI assistant or IDE. The tool uses Puppeteer’s headless Chrome to take screenshots quickly, but it also offers advanced options: responsive viewport sizing, configurable wait strategies for dynamic content, and automatic cleanup of browser instances to keep resource usage low. The result is a fast, deterministic screenshot that can be fed directly into an AI pipeline without additional preprocessing.

Key capabilities include:

• Fast capture: Leveraging Puppeteer’s efficient rendering to minimize latency.
• Vision‑optimized sizing: Automatic 1072 × 1072 resolution and tiling for full‑page support.
• Screencast recording: Capture a series of screenshots over time, exported as animated WebP for easy integration with video‑analysis models.
• Custom JavaScript injection: Run arbitrary scripts before capturing to manipulate the page or wait for specific states.
• No caching: Every request produces fresh content, ensuring up‑to‑date visual data.
• Resource efficiency: The server automatically shuts down idle browser instances after 60 seconds and closes pages after each screenshot, preventing memory leaks.

Real‑world use cases span automated visual testing, data collection for training vision models, and dynamic content analysis. For example, a QA engineer can instruct an AI assistant to “capture the current state of the login page” and receive a perfectly sized image ready for OCR or layout verification. A data scientist can generate a large dataset of page tiles to train a model that detects UI elements, while a content creator can quickly produce animated previews of web interactions for documentation.

Integration is seamless: the MCP server exposes a single tool with intuitive parameters (, viewport dimensions, wait strategies). AI assistants can call this tool in a conversation or script, receive the image data, and immediately pass it to vision APIs. Because the server handles all browser orchestration internally, developers can focus on higher‑level logic without worrying about headless browsers or image preprocessing.