# MAKO — Markdown Agent Knowledge Optimization

> Open protocol for serving LLM-optimized web content. 93% fewer tokens than HTML.

MAKO transforms web pages into structured, token-efficient Markdown that AI agents can read faster and smarter. It uses HTTP content negotiation (Accept: text/mako+markdown) so agents get optimized content while browsers get normal HTML.

## Key Features
- **93% token reduction**: A product page goes from 4,125 tokens (HTML) to 276 tokens (MAKO)
- **Compact Embedding Format (CEF)**: 512-dim embeddings in ~470 bytes via int8 quantization + gzip + base64url
- **Declared actions**: Machine-readable actions (add_to_cart, subscribe) with endpoints
- **Semantic links**: Internal, external, and API links with context descriptions
- **10 content types**: product, article, docs, landing, profile, listing, event, recipe, faq, custom
- **3 conformance levels**: Level 1 (file), Level 2 (HTTP), Level 3 (CEF)

## How It Works
1. Agent sends HEAD request with `Accept: text/mako+markdown`
2. Server responds with X-Mako-* headers (tokens, type, embedding)
3. Agent checks CEF embedding similarity against query
4. If relevant, agent sends GET for full MAKO content

## Links
- Spec: https://github.com/juanisidoro/mako-spec
- JS Library: https://github.com/juanisidoro/mako-js
- npm: https://www.npmjs.com/package/@mako-spec/js
- Website: https://makospec.vercel.app

## Install
npm install @mako-spec/js

## License
Apache-2.0