Education · WebMCP in 2026

What is WebMCP?

Web Model Context Protocol is a protocol that allows AI agents not just to find your website — but to act on it. Buy, book, search.

Open standardDeclarative HTMLWorks with ChatGPTWorks with Gemini
The problem

AI sees your site like a browser from the 90s.

ChatGPT and other models "read" pages like text browsers. They don't understand what's a product, a price, or how to buy — unless you tell them.

Without WebMCP
AI sees only raw HTML
AI doesn't know that 249 PLN is a product price
AI can't click "Add to cart"
AI doesn't know size availability
AI can't search the catalog
Your store disappears from AI recommendations
With WebMCP
AI understands and executes actions
"Find a dress size 38 under 300 PLN" → accurate results
"Add to cart" → AI performs the action
AI knows stock levels and variants
AI can search the product catalog
Your store appears in AI recommendations
How it works

Three components of WebMCP

WebMCP is a set of HTML and JavaScript conventions that together create an "interface" for AI agents — without rewriting your backend.

AI Agent
ChatGPT, Gemini, Copilot — asks a question or requests an action
WebMCP Layer
llms.txt + HTML attributes + Tools API — instructions for AI
Your website
Shop, service, app — executes the action or returns data
Declarative
Declarative WebMCP
You add data-mcp-* attributes to existing HTML. Zero JavaScript, zero backend. The ideal starting point.
index.html
<!-- Search form -->
<form data-mcp-action="search"
      data-mcp-description="Wyszukaj produkty"
      data-mcp-param-query="#search-input"
>
Implementation time: ~2h
Imperative
Imperative WebMCP
Register Tools via JavaScript API. Full control, error handling, backend integration. For complex actions.
cart.js
// Register "add to cart" tool
window.mcpTools.register({
  name: "add_to_cart",
  description: "Dodaj produkt",
  handler: async ({ id }) => { ... }
});
Implementation time: 4–6h
Capabilities

What AI can do on your website

Each "tool" is a separate action an AI agent can call. Here's what we most often implement for clients.

Tool / actionWhat AI doesDifficulty
search_products
Searches products by parameters (price, size, color)
Easy
add_to_cart
Adds a product to the user's cart
Medium
get_availability
Checks stock levels and available variants
Easy
check_delivery
Checks delivery options and times
Medium
contact_form
Fills and submits the contact form
Easy
book_appointment
Books an appointment, Calendly integration
Hard
llms.txt

The new robots.txt for AI

The /llms.txt file is an instruction for language models — how to understand your site and what they can do on it.

/llms.txt file in domain root
A short text file that AI fetches automatically. Like robots.txt, but for language models.
Human and AI readable
Markdown + YAML. Zero special syntax. Any developer can read it in 30 seconds.
You control what AI knows
You specify which tools are available, who you allow (GPTBot, PerplexityBot) and how to contact you.
auditai.cc/llms.txt
# llms.txt — instrukcja dla agentów AI
# auditai.cc/llms.txt

name: Sklep Mody Anna
description: Sklep z odzieżą damską. Rozmiary 34-46.
language: pl

tools:
  - search_products: Wyszukaj produkty wg parametrów
  - add_to_cart: Dodaj produkt do koszyka

allow: GPTBot, Googlebot-Extended, PerplexityBot
contact: [email protected]

Check if you have WebMCP

Our audit checks in 60 seconds whether your site is ready for AI agents — and tells you exactly what to fix, in what order.

Run free audit
60 secondsNo registration50 checkpoints