Pagewright – AI Browser Control & Scraping (MCP)

Pagewright connects your browser to AI agents through the Model Context Protocol (MCP). Any AI app that supports MCP can read pages, click, type and take screenshots in the browser you already use, with your existing logins. WHAT YOUR AGENT CAN DO • Read any page as a compact outline of headings, text and interactive elements, each with a short reference the agent can click or type into • Take screenshots of the visible area, the full page, or a single element • Click, type, pick options, press keys, hover, scroll, upload files and wait for content, using real input events that work with React, Vue and other modern apps • Open, switch and close tabs, and go back, forward or reload FIX FRONTENDS FASTER • Layout audit: finds horizontal overflow and the element causing it, content spilling out of its container, clipped text, elements covered by others, broken or stretched images, low color contrast, tiny text and tap targets, and unlabeled buttons • Inspect any element: box model, computed styles, and the CSS rules that apply, with file and line number • Test mobile, tablet and desktop screen sizes • Try a CSS fix live on the page, then check it with a screenshot before editing your code • Read console errors and network requests SCRAPE DATA • Get any page as clean markdown, a list of links, tables as JSON, or metadata (title, description, OpenGraph, JSON-LD) • Extract repeated items (products, listings, search results) into structured records • Follow "next page" buttons and infinite scroll, remove duplicates, and save to CSV or JSON • See the API requests a page makes and read their JSON responses PRIVATE BY DESIGN • Connects only to a server on your own computer (127.0.0.1). Websites cannot connect to it. • No analytics, no tracking, no accounts. The developer receives none of your data. • Controlled tabs show Chrome's debugging notice, and you can release any tab from the popup. SETUP (about a minute) 1. Install Node.js 18 or newer. 2. Run: npx -y github:zamansheikh/browser-mcp setup This installs the local MCP server and shows how to add it to your AI app. 3. Restart your AI app and ask it to run browser_status. Pagewright is open source (MIT): https://github.com/zamansheikh/browser-mcp Note: pages built into the browser (chrome:// pages and the Chrome Web Store) cannot be controlled. That is a Chrome restriction.