Until now, the conversation about visibility in AI has mostly been about whether language models know a brand and cite its content correctly. WebMCP \u2013 a proposed web standard \u2013 pushes that discussion a step further. What matters is no longer just whether an AI agent can read a page. What matters now is whether the page can tell the agent what it can do there, and how.<\/p>\n\n\n\n
The most important points at a glance:<\/p>\n\n\n\n
WebMCP is a proposed web standard meant to help sites expose structured tools to AI agents. Chrome’s documentation describes it as a way to expose tools through JavaScript and declarative HTML form annotations. A page no longer has to hope an agent will guess a button’s meaning from the look of the interface. Instead, it declares: here is the search box, here is the contact form, and here is the checkout. The agent gets an unambiguous description of the action and a schema for the input data. The result is fewer errors and faster task completion.<\/p>\n\n\n\n
WebMCP is still at an early stage, though. Chrome published its documentation on May 18, 2026, and the API may still change. It is not yet a standard supported by every browser.<\/p>\n\n\n\n
WebMCP makes the most sense wherever an AI agent uses a site the way it would use an app with clearly described functions. The paths most prone to errors stand to gain the most: bookings, configurators, checkouts, lead forms, and internal search. For a marketer, this is a new type of optimization. It’s no longer just about whether a human user understands the page, but also whether their agent knows what it can do there.<\/p>\n\n\n\n
WebMCP differs from MCP mainly in where it operates. MCP, or Model Context Protocol, connects AI agents to external systems, data, and processes, independent of any open page. WebMCP tools exist only in the open browser tab and disappear once it’s closed. So WebMCP doesn’t replace MCP \u2013 it solves a different problem. Here are the key differences at a glance:<\/p>\n\n\n\n The scope of WebMCP’s actions is set by the site owner. The agent only uses tools defined by the site, for example searching, filtering, or adding a product to the cart. It can change the state of the interface while doing so, but only within the permissions and context of the user. This does not mean an external model can freely edit the page. Poorly designed tools increase the risk of bad leads, while well-described ones shorten the path from intent to conversion.<\/p>\n\n\n\n The main value of WebMCP lies in reducing uncertainty. Today’s agent reads a page the way a human would, parsing text, layout, and form labels. That can be unreliable with non-standard components, dynamic filters, and multi-step paths. With a tool contract and a data schema, the agent has to guess less and completes tasks more reliably. The brand keeps control over UX and business rules throughout.<\/p>\n\n\n\n A WebMCP tool registers itself via You enable WebMCP in Chrome manually, as an experimental feature. Local testing requires Chrome Canary or the Dev Channel, and an origin trial for stable Chrome is planned starting with version 149. Open BeeClear WebMCP AI Visibility is a free plugin that helps you implement WebMCP integration without writing your own code and without external services. It lets AI agents see your site’s content structure, forms, and navigation as ready-made tools.<\/p>\n<\/div>\n\n\n\nArea<\/th> MCP<\/th> WebMCP<\/th><\/tr><\/thead> Main use case<\/td> Backend data, systems, and processes<\/td> Interaction with the currently open page<\/td><\/tr> Lifecycle<\/td> Persistent server or service<\/td> Tools available during a page visit<\/td><\/tr> Context<\/td> Global, independent of the UI<\/td> Browser, DOM, and page state<\/td><\/tr> SEO\/marketing perspective<\/td> Distributing data and functions to agents<\/td> Making it easier for agents to take action on the page<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n How broad a scope of action does WebMCP allow?<\/h2>\n\n\n\n
What is the main value of WebMCP?<\/h2>\n\n\n\n
How does a WebMCP tool work technically?<\/h2>\n\n\n\n
document.modelContext.registerTool()<\/code>, providing a name, description, JSON schema, and an asynchronous execute<\/code> function. As of Chrome 150, navigator.modelContext<\/code> is deprecated. When an agent lands on a page, it fetches the list of tools through getTools()<\/code> and reads their contract the way it would read function descriptions in function-calling systems. That’s why the quality of the description<\/code> field and the precision of the inputSchema<\/code> determine effectiveness. A weak description makes the model guess or simply ignore the tool.<\/p>\n\n\n\nif (document.modelContext) { document.modelContext.registerTool({ name: 'search_products', description: 'Search products by keyword and optional category.', inputSchema: { type: 'object', properties: { keyword: { type: 'string', description: 'Search keyword.' }, category: { type: 'string', description: 'Optional product category slug.' } }, required: ['keyword'] }, execute: async ({ keyword, category }) => { const params = new URLSearchParams({ s: keyword }); if (category) params.set('product_cat', category); window.location.href = `\/sklep\/?${params.toString()}`; return `Product search opened for: ${keyword}`; } }, { exposedTo: [window.location.origin] }); }<\/code><\/pre>\n\n\n\nHow do you enable WebMCP in Chrome?<\/h2>\n\n\n\n
chrome:\/\/flags\/#enable-webmcp-testing<\/code>, set it to \u201cEnabled,\u201d and restart the browser. Without the flag, the document.modelContext<\/code> object doesn’t exist, so your implementation should check for its availability. For diagnostics, use the WebMCP panel in Chrome DevTools and the Model Context Tool Inspector extension.<\/p>\n\n\n\nGet your WordPress site ready for AI agents<\/h2>\n\n\n\n