GlobalDex, built by Said Borna — a Swedish DevOps engineering student at Chas Academy — runs 34 deterministic compliance checks across five categories: structure, metadata, accessibility, discoverability, and WebMCP support. Any scanned URL gets a 0–100 score and letter grade. Results then go to Anthropic's Claude, which identifies the site's industry and writes a two-to-three sentence assessment of strengths, weaknesses, and recommended fixes. The platform is free, requires no sign-up, and ships as a web scanner, CLI, GitHub Action, Chrome extension, and REST API.
The headline technical claim is WebMCP detection. Web Model Context Protocol is a browser API that lets websites explicitly declare structured tools for AI agents, rather than leaving agents to infer page structure and scrape form interactions. Borna compares it to a REST API contract, but browser-native. He says GlobalDex is the first scanner to check for it. WebMCP is targeted for Chrome 146, though the spec does not yet appear on the W3C standards track and Google has not published a formal intent-to-ship; its status sits closer to a Chrome experiment than a ratified web standard. That context matters for developers weighing how much to invest in early compliance.
The CI/CD integration is the most practically useful piece for engineering teams. Configure a minimum score threshold and the GitHub Action exits with code 1 if the site falls short, blocking a deployment the same way a failed accessibility audit would. GlobalDex also indexes over 50,000 domains sourced from the Tranco top-1M list and supports embeddable score badges.
What the tool actually bets on is that autonomous browsing agents — from Google, OpenAI, Anthropic, and others — will increasingly favor sites that <a href="/news/2026-03-14-optimizing-web-content-for-ai-agents-via-http-content-negotiation">surface structured, machine-readable intent</a> over sites that require agents to reverse-engineer every interaction. Whether WebMCP becomes the mechanism that delivers that or gets superseded by something else, the underlying pressure on web developers to think about agent legibility is real. Borna is early, possibly too early — but the tooling to measure that readiness has to exist before anyone can act on it.