Introducing MAKO

The web wasn't designed for AI agents. Every time ChatGPT, Perplexity, or a shopping assistant visits a website, it downloads navigation bars, cookie banners, ad scripts, and thousands of lines of markup — just to find a product name and price.

The result? 4,000+ tokens consumed before the agent reaches the actual content. For JavaScript-rendered SPAs, the situation is even worse: the agent sees an empty <div id="root"></div> and nothing else.

The Problem

Consider a typical e-commerce product page. A human visitor sees a clean layout with a product image, title, price, and an "Add to Cart" button. An AI agent sees this:

• 181 KB of raw HTML
• Navigation with 47 links
• 3 cookie consent scripts
• Inline CSS for 200+ components
• The actual product data buried somewhere in the middle

This is 93% noise, 7% signal.

The SPA Black Hole

For modern JavaScript applications (React, Vue, Angular, Next.js CSR), the situation is even worse. When an AI agent requests a page, the server responds with something like:

<div id="root"></div>
<script src="/bundle.js"></script>

The actual content — products, articles, prices, descriptions — is rendered client-side by JavaScript. AI agents don't execute JavaScript. They see an empty page. Your entire site is invisible.

Server-side rendering (SSR) and static generation help, but even then the HTML is bloated with hydration markers, embedded state, and framework artifacts. A Next.js SSR page still sends 5-10x more tokens than necessary because it includes the full React tree alongside the actual content.

The web is split: legacy sites send too much noise, modern SPAs send nothing at all. Neither format is designed for the audience that's growing fastest.

What MAKO Does

MAKO adds a structured, AI-optimized layer to any website using standard HTTP content negotiation. When an AI agent sends Accept: text/mako+markdown, the server responds with a clean MAKO document instead of raw HTML:

---
mako: "1.0"
type: product
entity: "Wireless Headphones Pro"
tokens: 276
language: en
updated: "2026-02-20T10:00:00Z"
---

The same URL, the same server — just a different response for a different audience.

Key Design Decisions

Content negotiation over new endpoints. We chose the Accept header pattern because it requires zero URL changes. Every existing page can serve MAKO without creating duplicate routes.

Markdown over JSON. LLMs are trained on markdown. A well-structured markdown document with YAML frontmatter is more token-efficient and more naturally readable by language models than equivalent JSON.

10 content types. Product, article, docs, landing, listing, profile, event, recipe, FAQ, and custom. Each type tells the agent exactly what structure to expect.

Declared actions. Instead of hoping the agent finds your "Add to Cart" button in the HTML, MAKO declares machine-readable actions with endpoints and parameters.

The Numbers

In our benchmarks across 50+ real-world pages:

That's not a marginal improvement. It's the difference between an agent that can process 3 pages per context window and one that can process 45.

Getting Started

MAKO is available today:

• WordPress: Install the mako-wp plugin — activate and it works with WooCommerce, Yoast, and ACF
• Any stack: Use @mako-spec/js — TypeScript SDK with Express middleware
• Validate: Use @mako-spec/cli to validate your MAKO files

The protocol is open source under Apache 2.0. The full specification is available at makospec.vercel.app/docs.

What's Next

We're working on MAKO Score — an auditing tool that measures how AI-ready any website is across four dimensions: Discoverable, Readable, Trustworthy, and Actionable. Check your site at makospec.vercel.app/score.

The web is gaining a new audience. MAKO helps you speak their language.