What Is WebMCP? Making Websites Ready for AI Agents

Websites have generally been designed around one assumption: a human being is sitting in front of a browser.

Pages are rendered visually. Navigation menus help people find information. Buttons tell people where to click. Forms give people fields to complete. Behind all of this is an interface specifically designed for human interaction.

AI agents are beginning to challenge that assumption.

Increasingly, users will delegate tasks to AI agents that browse websites, find information, compare options, fill out forms, make purchases, schedule appointments, and perform other actions on their behalf. The problem is that today's websites weren't designed for these machine visitors.

WebMCP is an emerging standard designed to change that.

What Is WebMCP?

WebMCP is a proposed web standard that enables websites to expose structured tools directly to AI agents.

The WebMCP specification is being developed through the W3C Web Machine Learning Community Group, with editors from Google and Microsoft. The specification describes WebMCP simply as an API that "enables web applications to provide JavaScript-based tools to AI agents."

Instead of forcing an AI agent to interpret a website visually and simulate what a person would do, a website can explicitly tell an agent what capabilities are available.

For example, an e-commerce website might expose tools for searching products, filtering results, checking inventory, or adding an item to a cart. A travel website could expose tools for searching flights or hotels. A healthcare website might expose an appointment scheduling tool.

Google describes WebMCP as a way to expose structured tools that improve the performance and reliability of agent interactions with websites. Tools can define explicit inputs using JSON Schema, giving agents a much more deterministic way to interact with a site.

That is fundamentally different from today's browser automation.

From Screen Scraping to Structured Interaction

Consider an AI agent trying to schedule an appointment on a traditional website.

Without WebMCP, the agent may need to examine the DOM or rendered page, identify the appointment button, click it, determine which fields correspond to date and location, populate those fields, click another button, interpret the resulting page, and continue through the workflow.

Every step requires interpretation. Change the website layout or rename a button and the agent may have trouble completing the task.

WebMCP provides another approach. The website can explicitly expose something conceptually equivalent to a scheduleAppointment tool and define exactly what information that tool requires.

The agent no longer needs to figure out how a human would use the website. It can understand what the website is capable of doing. And it's a major architectural shift.

Websites begin to have two interfaces: a visual interface optimized for humans and a structured, semantic interface that AI agents can understand.

WebMCP Is Not MCP

Despite the name, WebMCP is not simply MCP running inside a website.

The Model Context Protocol (MCP) provides a standardized way for AI applications and agents to connect with external systems, data sources, and tools. WebMCP is focused specifically on interactions between browser-based agents and websites.

Google describes MCP as primarily backend-oriented and WebMCP as frontend-oriented. MCP services can be available independently of a website, while WebMCP tools exist in the context of the page and browser session.

The technologies are therefore complementary.

An enterprise might expose backend content and services through an MCP server while also exposing appropriate website functionality through WebMCP. The first makes enterprise capabilities available to AI applications broadly. The second makes the live website itself easier and more reliable for browser-based agents to use.

For organizations preparing for an agentic web, both are likely to become important.

Why Structured Content Matters for WebMCP

WebMCP solves an important part of the machine experience problem: how an AI agent acts on a website.

But actions are only part of the experience. Before an agent can act, it frequently needs to understand.

Imagine an agent shopping for a laptop. It needs to understand products, prices, specifications, availability, warranties, shipping policies, and potentially hundreds of other pieces of information before deciding what action to take.

If that information exists primarily as presentation-oriented HTML or large blobs of rich text, the agent has to infer its meaning.

Structured content provides a much better foundation.

Instead of seeing: "The Enterprise Plan includes advanced security, SSO, unlimited projects..." an application can work with explicitly modeled fields such as plan name, price, features, user limits, support level, and security capabilities.

That distinction becomes increasingly important as machines become first-class consumers of website content.

Structured content separates information from presentation. Content is modeled as types, fields, metadata, and relationships rather than being embedded entirely within the visual representation of a web page. This makes the information easier to query, retrieve, reuse, and interpret programmatically.

In other words: WebMCP makes website actions machine-readable. Structured content makes website information machine-readable.

The agentic web needs both.

Structured Content Creates Better Agent Experiences

This also suggests that simply adding WebMCP tools to an existing website isn't the complete solution.

An agent might know that a website offers a findProduct tool, for example, but the quality of the experience still depends heavily on the quality and structure of the information behind that tool.

Well-modeled content provides explicit semantics. A product has a name, description, category, specifications, images, compatibility information, and relationships to other products. An event has a location, start time, end time, speaker, capacity, and registration information.

That structure reduces ambiguity.

It also makes it much easier for developers to create WebMCP tools because the underlying information architecture already exists in machine-readable form.

This is where the CMS becomes an important component of WebMCP and the broader agentic web architecture.

CrafterCMS and the Agentic Web

CrafterCMS is particularly well suited to this emerging model because structured content and API-driven delivery are fundamental parts of the platform.

Content teams can define content models for products, articles, services, locations, events, documentation, or virtually any other domain object. Content is stored independently of its presentation so the same information can be rendered into a traditional website while also being consumed programmatically.

CrafterCMS then makes that structured content available through multiple interfaces, including REST APIs, GraphQL, JavaScript SDKs, search APIs, and custom APIs.

Its native GraphQL implementation is especially interesting in this context. CrafterCMS automatically generates a GraphQL schema based on the site's content models and updates that schema as those models change.

That means the same content model used by authors to manage an experience can become a predictable machine-readable representation of the organization's content.

Developers can then build WebMCP tools into the website using JavaScript or declarative HTML while retrieving the underlying structured content from CrafterCMS.

CrafterCMS's extensibility also means developers aren't restricted to content retrieval. Custom APIs and application logic can integrate CMS content with commerce platforms, CRM systems, search, authentication, inventory, and other enterprise services.

The result is an architecture capable of serving both information and actions to AI-powered experiences.

From Headless CMS to Agentic CMS

WebMCP is part of a much larger change taking place on the web.

For decades, content management systems primarily served human authors creating content for human visitors. Headless CMS architecture expanded that model by separating content from presentation so the same content could serve websites, mobile applications, digital signage, voice assistants, and other channels.

AI agents introduce another channel — but this one is different.

Agents don't merely consume pages. They interpret information, make decisions, and take actions.

The next generation of CMS architecture is agentic, and needs to consider four participants: human authors, AI authors, human visitors, and AI visitors.

For AI visitors, structured content provides the information layer. APIs, MCP, and technologies such as WebMCP provide the interaction layer.

This is why the move toward structured content is about much more than traditional omnichannel publishing. Content that has been separated from presentation, modeled semantically, and exposed through APIs is inherently better prepared for machines to understand and use.

Preparing Your Website for WebMCP

WebMCP is still an emerging technology. The specification is currently a W3C Community Group draft, and browser implementations are evolving. Google currently describes the technology as under active discussion and subject to change.

But the direction is significant.

The web is moving from an environment where AI agents must reverse-engineer interfaces designed for humans toward one where websites explicitly expose their capabilities to machines.

Organizations don't need to wait for WebMCP to become ubiquitous to prepare for that future.

The first step is ensuring that website content is properly modeled and structured rather than trapped inside presentation-oriented pages. The next is making that content available through clean APIs and search interfaces. From there, organizations can expose increasingly sophisticated capabilities through MCP, WebMCP, and other emerging agent protocols.

CrafterCMS provides that foundation today: structured content, flexible content modeling, API-first delivery, native GraphQL, powerful search, and an extensible development platform.

The web was originally designed for people to navigate. The next web will increasingly be designed for people and their agents.

WebMCP gives those agents a better way to act. Structured content gives them something better to understand.

Learn more about structured content in our blog post: Structured Content in a Headless CMS