WebMCP is a new browser standard that makes websites agent-ready. Find out what it means for eCommerce brands, how it works, and what you can do to prepare.

Right now, AI agents are visiting your website.

They are browsing products, filling forms, and completing transactions on behalf of humans. But what they’re doing is, ultimately, rudimentary guesswork.

This is because the web was built for human eyes and human hands. When an AI agent lands on your product page, it has to screenshot the screen, parse the HTML, guess what the filter dropdowns do, and hope the date picker accepts the format it chose. It is slow, it burns tokens, and it breaks every time someone updates the frontend.

But that is going to change with the help of WebMCP.

For brands selling online, you’ll want to pay attention. Not because you need to rebuild anything today, but because those who prepare for this early are going to have an advantage as agentic commerce becomes the norm.

Here is what WebMCP is, how it works, and what you can do about it right now.

What is WebMCP?

Web Model Context Protocol (WebMCP) is a proposed web standard that lets websites declare their capabilities as structured tools that AI agents can call directly.

This means that your website can say, “I have a tool called 'search-products'. It needs a query, a price range, and a category. Call it and I will return matching results.” The agent calls the tool, gets structured data back, and moves on without needing to guess which button does what.

Alex Nahas, a backend engineer at Amazon, built the precursor to WebMCP in early 2025. Amazon had spun up thousands of internal MCP servers after Anthropic released the Model Context Protocol (MCP), and the result was a mess. Thousands of tools crammed into single context windows, and an OAuth 2.1 authentication requirement that almost nobody had implemented internally.

Nahas's idea was a simple one. The browser already solved the authentication problem. It had session cookies, single sign-on, and role-based access controls. He built MCP-B (the ‘B’ stands for browser), integrating the MCP SDK into client-side JavaScript and using postMessage to communicate with a Chrome extension.

When Google's Chrome team and Microsoft's Edge team saw the approach, they recognised it solved a problem they had been circling independently. The three groups joined forces through the W3C, and the result is WebMCP.

How AI agents interact with sites today

To understand why WebMCP matters, let’s understand what it replaces. There are two ways AI agents currently work with websites, and both have problems.

The first is visual interpretation. The agent takes a screenshot, sends it to a vision model, and tries to identify buttons, fields, and navigation elements from the image. This works (most of the time), but it is slow and expensive. Every action requires a new screenshot, a new inference call, and a new set of guesses about what each element does.

The second approach is DOM parsing. The agent reads the raw HTML and tries to figure out meaning from tags, class names, and page structure. This is faster than screenshotting, but fragile. A developer renames a CSS class, restructures a component, or moves a form field, and the agent's logic breaks. It also requires a developer to have the foresight to name classes with some coherence.

Both methods force the agent to reverse-engineer a website that was designed for humans, which leads to complications.

WebMCP removes that step entirely. The website tells the agent what is available.

What WebMCP is not

Now that we know what WebMCP is, let’s understand what it isn’t.

WebMCP is not MCP. It shares MCP's conceptual model (tools with schemas that agents can call), but it does not follow the JSON-RPC specification. The W3C working group chose not to couple the browser standard too tightly to Anthropic's protocol. So it’s best to think of WebMCP as "MCP-inspired APIs for the browser" rather than a direct port.

WebMCP is not a replacement for backend integrations. Chrome's own documentation from March 2026 says that WebMCP and MCP address different needs. Your existing APIs, ERP integrations, and data pipelines stay exactly where they are. WebMCP simply handles the browser-side interaction layer.

WebMCP only covers tool calling. MCP has concepts for "resources" (structured data sources) and "prompts" (template interactions). WebMCP has no plans to implement either. It does one thing: let websites expose callable tools.

And WebMCP is not production-ready… yet. It sits behind a feature flag in Chrome Canary. The specification has already changed since launch (the provideContext and clearContext methods were removed in March 2026), and production readiness across Chrome and Edge is expected in the second half of 2026.

Universal Commerce Protocol (UCP)

Let’s also take a moment to talk about another big component rising up the agentic ranks.

UCP is an open standard co-developed by Google and Shopify, with endorsement from more than 20 big partners, including Visa, Mastercard, Etsy, Wayfair and Walmart. It is designed to standardise the full commerce transaction journey from product discovery and cart management through to checkout, payments, and order management.

Where WebMCP lets an agent interact with your website's interface, UCP handles the actual money. It lets agents complete checkout on a customer's behalf, apply discount codes, select shipping options, and process payment through standardised flows.

UCP is already rolling out on Google AI Mode and the Gemini app with select US retailers. Shopify merchants can manage these integrations centrally through what Shopify calls Agentic Storefronts.

Give it a few years (maybe not even that long) and most sites will use a combination of WebMCP and UCP to give agents free range of the discovery and purchase journey, without humans needing to so much as peep at the process.

Why should businesses care about WebMCP?

So if WebMCP isn’t ready yet, why should we care? Because the best time to prepare is before you need to. Let’s stop talking about the technology and start talking about what it means for your business.

The agentic browser market went commercial in early 2026. Google's Chrome “auto browse” lets users delegate browsing tasks to Gemini directly inside the browser. OpenAI's Atlas can open tabs, navigate airline websites, extract pricing, and present comparisons. Perplexity's Comet completed its cross-platform rollout in March 2026.

These products are built for real tasks. They are acting on behalf of users, and like any user, they want to use sites that work with them, not against them.

The sites that make it easy for these agents to complete tasks will capture the traffic. The ones that force agents to guess will get skipped for competitors that do not.

There’s also the time element of it. When smartphones started outselling desktops, sites that adopted responsive design early won the distribution game. They ranked better in mobile search. They converted more mobile traffic. Late movers spent the next few years catching up while traffic shifted around them.

WebMCP adoption is going to follow the same pattern, and implementation isn’t difficult. The declarative API (more on this later) requires two HTML attributes on your existing forms. No re-architecture, new backend, or in-depth jiggery-pockery.

What WebMCP means for DTC businesses

For direct-to-consumer brands, WebMCP reduces friction at the points that matter most to conversion and retention.

Product discovery, checkout, subscription management, and returns can all be exposed as structured tools that agents call directly, instead of trying to navigate your UI. When an agent can search your catalogue, manage a cart, apply a discount code, and complete checkout without guessing its way through each step, there are fewer places for the journey to break down.

For brands where acquisition costs are rising, that reduction in friction is margin.

What WebMCP means for B2B businesses

B2B is where WebMCP could have the most practical impact.

Quote requests, procurement workflows, approval chains, and large catalogue search can all be exposed as callable tools, which means a buyer's agent can interact with your site reliably instead of struggling with a custom interface it was not designed for.

One detail that matters here specifically. WebMCP inherits the user's browser session, so tools automatically respect logged-in pricing, stock visibility, and account-specific terms with no separate auth layer to build.

For businesses working to shift trade customers from phone and counter ordering to digital self-serve, structured tools lower the barrier for buyers who find complex web interfaces intimidating.

How does WebMCP work?

There are two ways to make your website agent-ready with WebMCP: a declarative approach using HTML and an imperative approach using JavaScript.

The declarative API

This is the low-effort option, and for many eCommerce sites it will cover the most common use cases.

If your site already has HTML forms (such as search bars, enquiry forms, login pages, quote request forms et cetera), you can make them agent-compatible by adding two attributes to the form tag:

toolname (the tool identifier)
tooldescription (a plain-language description of what the tool does).

It’s likely that more attributes will come later down the line, but for now, these two are the main attributes to consider.

If we take these and apply them to an example – say in a search bar – here’s what that could look like:

<form toolname="search-products" tooldescription="Search the product catalogue by keyword, category, and price range" toolautosubmit="true">

The remainder of the form would then continue as intended.

When an agent visits the page, the browser reads these attributes and translates the form into a structured tool definition that the agent can understand and call. The agent sees the tool name, reads the description, knows what parameters are available, and can fill in the form with the right values.

And yes, your eyes haven’t deceived you; there is also the optional toolautosubmit attribute we’ve added in there, which tells the browser to submit the form automatically once the agent fills it in. Without it, the agent fills the fields and waits for the user to confirm.

That is really all it takes for the declarative approach. Two attributes on the form tag, plus optional descriptions on individual inputs.

The imperative API

For interactions that go beyond form submission, there is a second approach.

Developers register tools programmatically using JavaScript, defining a name, a description, the expected inputs, and a function that runs when the agent calls the tool.

This is the route for anything dynamic, such as filtered product searches, cart operations, or B2B pricing lookups that depend on the logged-in account. The developer writes the logic once, and any compatible agent can call it.

Two details worth knowing even if you are not writing the code yourself.

First, tools can be flagged as read-only. A stock check or a product search does not change anything, so the browser can skip the user confirmation step and let the agent run it immediately. Write operations, like adding to cart or placing an order, require the user to confirm before they execute. Your development team decides which tools get which treatment.

Second, tools can appear and disappear based on context. A product detail page registers an add-to-cart tool and a check-availability tool. The cart page registers update-quantity and apply-discount. The checkout page has its own set. When the user navigates away, those tools unregister. The agent only ever sees what is relevant to the page it is on, which keeps interactions focused and reduces the chance of errors.

The three layers for agentic commerce

WebMCP does not work in isolation. It sits within a broader stack of protocols, and understanding how they fit together helps you plan what to prioritise:

Structured data (Schema.org markup) helps agents discover and understand your products. Product names, prices, availability, reviews, shipping details. Without this foundation, agents cannot find you. Most eCommerce sites have some of this in place already, though coverage and consistency vary.

WebMCP helps agents interact with your site. Search for products, filter results, check stock, manage carts, submit enquiries, track orders. This is the interaction layer that sits between discovery and transaction.

UCP and backend MCP handle the commercial transaction and operational backend. Checkout, payment processing, order management, inventory systems, ERP connections. This is where the money moves and the operations run.

Most sites have the first layer partially in place. The second and third are where preparation starts now.

How to prepare your website for WebMCP

Want to make sure your website is prepared for WebMCP? Here is a clear approach that starts with work you can do that builds toward full agent-readiness.

Step one: audit and clean up

This costs nothing and improves your site regardless of whether WebMCP reaches your customers this year or next.

List the five to ten most important actions on your site. Product search, enquiry forms, quote requests, checkout flows, account login, order tracking, return initiation. Whatever you decide, these are the actions that will become WebMCP tools.

Then look at those forms. Are the labels clear? Are the input types correct? Do they validate properly and redirect predictably? Clean, well-structured HTML is the raw material WebMCP works with. The same form hygiene that helps screen readers and search engines is exactly what agents need.

Review your structured data at the same time. Product and Organisation markup should be consistent across all pages. Gaps in structured data mean gaps in visibility to AI systems.

Step two: add declarative WebMCP attributes

Add toolname and tooldescription to your highest-value forms. Product search, your main enquiry form, and account login. Keep it to two attributes per form.

Write tool descriptions as if you were writing conversion copy. The name and description directly determine whether an agent chooses your tool over a competitor's. Use clear verbs. Be specific about what the tool does and what inputs it expects.

Then test. Install Chrome Canary, enable the WebMCP flag, and use the Model Context Tool Inspector extension to check that your tools appear and execute correctly.

Step three: deeper integration

When Chrome moves WebMCP to its stable release (expected mid-to-late 2026), the imperative API becomes worth investing in.

This means registering JavaScript tools for complex workflows. Product search with dynamic filters, real-time stock checks by location, cart management with validation, account-specific B2B pricing lookups, and quote configuration for trade customers.

The specifics depend on your platform and customers. Use contextual registration to keep things focused. Register tools when the relevant page loads, remove them when the user navigates away. The agent should only ever see the tools that apply to where the user is right now.

Plan how the full protocol stack connects for your operation. Structured data for discovery. WebMCP for site-level interaction. Backend MCP for inventory, ERP, and operational systems. UCP for checkout and payment if you are on a compatible platform. These layers are complementary, and mapping how they fit your setup is worth doing before the spec goes stable.

With these 3 steps, you can help progress your website to the next level of eCommerce.

Key steps to remember about WebMCP:

WebMCP is a proposed browser standard that lets websites declare their capabilities as structured tools AI agents can call directly. It was published as a W3C draft in February 2026 and is available behind a feature flag in Chrome 146 Canary.
WebMCP reduces friction across product discovery, checkout, quote requests, and B2B procurement by giving agents a reliable way to interact with your site instead of guessing their way through your UI.
Implementation starts small. The declarative API requires two HTML attributes on existing forms. The imperative API covers more complex workflows through JavaScript, but is worth waiting for until the spec stabilises in late 2026.
The preparation work is not wasted. Cleaner forms, consistent structured data, and well-defined site actions improve SEO, accessibility, and conversion right now. They also happen to be exactly what WebMCP builds on.

Velstar: an AI-first agency

WebMCP is early, but the preparation work is not wasted. Cleaner forms, better structured data, and well-defined site actions improve SEO, accessibility, and conversion today. They also happen to be the foundation WebMCP builds on.

If you want to get your website ready for WebMCP, Velstar is here to help. We help mid-market eCommerce brands build sites that perform now and are ready for what comes next.

If you want to understand how agent-readiness fits into your platform and performance roadmap, get in touch today.

WebMCP: What It Is, How It Works, and What Businesses Need to Do