The Agent-Readable Website Checklist Built by Yoast's Founder

Most SEO advice was written for search engines, not for the AI agents that now read, summarize, and act on web pages on a user's behalf. An agent-readable website checklist called specification.website tries to close that gap. It's a free, open-source spec covering HTML structure, accessibility, security, and SEO, with a dedicated section on what it takes for a page to be legible to an AI agent rather than just a human or a crawler. The project comes from Joost de Valk, the founder of Yoast SEO, and it's built as a reference document a person applies themselves, not a tool that scans a URL and hands back a score.

What This Actually Is

specification.website is a static reference site (source at github.com/jdevalk/specification.website) organized into ten categories: Foundations, SEO, Accessibility, Security, Well-Known URIs, Agent Readiness, Performance, Privacy, Resilience, and Internationalisation. Each of the 96 spec pages covers one requirement, for example a <title> tag or a security.txt file, with a status of Required, Recommended, Optional, or Avoid, plus a "why it matters," "how to implement," "common mistakes," and "verification" section, and at least one cited source.

That's the part worth being precise about. This is not an automated auditor. There's no crawler that visits a site and returns a pass or fail. It's a written specification and an accompanying flat checklist (at specification.website/checklist/) that a developer, or an AI agent following the checklist, works through by hand. The repo also ships a small Cloudflare Worker MCP server that exposes the spec's content, search, and category listings to MCP-aware clients like Claude Desktop, plus an audit_url prompt that builds an audit plan against a target URL. That prompt assembles a checklist-driven plan for an agent to work through, it does not run the scan itself. The distinction matters for anyone expecting a one-click grade.

Verifying the Authorship

The GitHub authorship checks out directly. The repository is owned by the account jdevalk, whose public GitHub bio reads "Internet entrepreneur... founder of @Yoast," and whose linked blog and Twitter handle (@jdevalk) match Joost de Valk's known identity as the founder of Yoast SEO, one of the most widely installed WordPress SEO plugins. The site's own README and the checklist page's Twitter card both attribute the project to @jdevalk. The credibility angle here isn't hype: someone who spent over a decade building tooling around what search engines actually reward has a reasonable claim to writing a spec about what makes a site legible to the next generation of automated readers.

What's Inside the Agent Readiness Category

The category most relevant to this site's readers covers things like llms.txt and llms-full.txt (machine-readable summaries of a site's content), structured data that agents can parse without rendering JavaScript, MCP and tool discovery conventions, and an ai-catalog.json file under /.well-known/ for agentic resource discovery. specification.website itself implements most of what it recommends: it ships its own llms.txt, llms-full.txt, per-page Markdown endpoints, and an MCP server, so the spec is also a working example of the practices it describes.

The rest of the categories are less novel but still useful as a single sourced reference: WCAG-aligned accessibility rules, security headers and /.well-known/security.txt, Core Web Vitals and caching guidance, and internationalisation basics like hreflang and text direction. None of this is new information on its own. What's different is having it collected in one place with citations back to the WHATWG HTML Living Standard, WCAG 2.2, IETF RFCs, and Google Search Central, rather than scattered across a dozen blog posts of varying quality.

Licensing and How to Use It

The code (the Astro-based site and the MCP worker) is MIT licensed. The spec content itself carries a second license, Creative Commons Attribution 4.0, meaning the written pages can be reused and adapted, including commercially, with attribution. Both checked out against the actual LICENSE file in the repo, matching the badges on the README.

Practically, there are two ways to use this. A team can work through the flat checklist manually, treating Required items as a baseline and Recommended items as a backlog. Or, for anyone already running an MCP-connected agent, the audit_url prompt gives the agent a structured plan to check a live page against the spec's Required and Recommended items, though the agent still has to do the actual fetching and comparing, the MCP server only serves the spec's content and the plan, it does not host a validator that runs the check for you.

Current Traction and What It Means

As of today, the repository sits at 830 stars and 48 forks, up slightly from the 826 stars noted at scan time. It was created in late May 2026, so it has under three months of history, and it saw a code push as recently as August 12. There's no long velocity track record to point to yet, which is why this pick rests on authorship credibility and the quality of the sourcing rather than a measured growth curve. Seven open issues suggest an active, if still small, contributor base working through edge cases.

Should You Use This

For anyone building or auditing a website with an eye toward both search visibility and agent readability, this is a genuinely useful reference to bookmark, not install. It won't scan a site for you, and it isn't a plugin or a service; it's a spec and a checklist, backed by real standards and written by someone with a credible track record in the exact problem space. The honest way to use it is the way the author frames it: read the relevant category, apply what's Required, decide deliberately on what's Recommended, and skip the parts explicitly marked Avoid. Anyone expecting an automated grade instead of a checklist to work through will be disappointed; anyone who wants a single sourced reference for what "agent-readable" concretely means will get real value from it.