WebMCP: Making your website ready for AI agents

For years, AI assistants that interact with the web have relied on an approach that can only be described as digital impersonation: take a screenshot of the page, feed it to a vision model, guess where to click, click, and repeat.

It is slow, fragile, and consumes a ridiculous amount of compute just to fill in a form that has <input name="email"> written in plain HTML.

There is a better way. It is called WebMCP, and it is one of the most interesting developments happening quietly at the intersection of browser standards and AI tooling right now.

In this post, we will explain what WebMCP is, why it matters for the projects we build, and how you can start implementing it today. We will keep the theory short and the code examples plentiful, because that is how we prefer to learn at MarsBased.

What is WebMCP?

WebMCP (Web Model Context Protocol) is a proposed web standard that allows websites to explicitly declare their capabilities in a structured form. This allows AI agents to call functions directly without needing to interpret the UI visually.

Instead of an agent scraping your DOM and guessing what a "Submit" button does, your website tells it: "Here is a tool called submit_contact_form, here is what it expects as input, and here is what it returns." The agent calls the tool, your application runs it, and you're done.

"WebMCP for websites is similar to what responsive design was for mobile: an incremental path to being ready for the next generation of users."

The standard was formally accepted as a W3C Community Group deliverable in September 2025 and received its first browser implementation in Chrome 146 (February 2026), driven by a joint effort from Google and Microsoft. It is currently behind a flag in Chrome Canary, with broader native support expected in the second half of 2026.

The responsive design parallel

When mobile hit the mainstream, teams did not rebuild their apps from scratch, they added responsive breakpoints and called it a day. WebMCP offers the same incremental path to agent-readiness: annotate your forms, register your key actions, and your web application becomes operable by an AI without a full re-architecture.

Why should you care?

Current AI browser agents work through a tedious loop: screenshot, vision analysis, identification, and execution. This leads to operations that take 5–10 seconds with a 15–20% error rate.

With WebMCP, those same operations run in 1–2 seconds with near-zero errors.

As AI agents become a significant source of web traffic not just bots crawling for content, but agents performing tasks on behalf of users, websites that are not agent-ready will simply be harder to use. Early adoption of WebMCP today is roughly analogous to optimizing for mobile search in 2012: not strictly necessary yet, but the teams that move early tend to dominate later.

For us at MarsBased, building primarily with React, Ruby on Rails, Node, and Python, WebMCP fits naturally into the form-heavy, data-driven applications we deliver.

Two APIs, one standard

WebMCP gives you two ways to make your application agent-ready. They are complementary, not competing.

1. The declarative API (HTML attributes)

This is the low-effort entry point. You add a handful of attributes to your existing <form> elements, and any WebMCP-capable agent can discover and use them automatically. No JavaScript required.

• toolname: A unique identifier for the tool.
• tooldescription: A plain-English explanation of what the form does.
• toolparamdescription: Context for each field (added to inputs, selects, or textareas).
• toolautosubmit: (Optional) Tells the agent it can submit automatically after filling.

Example:

<form
  toolname="contact_form"
  tooldescription="Send a contact request with your name, email, and message"
  toolautosubmit="true"
  noValidate
>
  <input
    type="text"
    name="name"
    placeholder="Your name"
    toolparamdescription="Full name of the person sending the request"
  />
  <input
    type="email"
    name="email"
    placeholder="your@email.com"
    toolparamdescription="Email address for follow-up communication"
  />
  <textarea
    name="message"
    toolparamdescription="The message body — be as detailed as needed"
  ></textarea>
  <button type="submit">Send</button>
</form>