Although artificial intelligence (AI) achieves impressive performance in complex tasks like code generation or passing legal exams, its ability to shop online autonomously remains limited. Most current agents designed for online shopping rely on analyzing the HTML structure of websites. They attempt to guess the function of individual elements, clicking and proceeding through trial and error. This approach is slow, resource-intensive (tokens), and moreover, fragile – even a minor change in page design can easily disrupt it.
The solution is WebMCP (Web Machine Comprehensible Pages) – a proposed browser standard from Google that provides AI agents with structured tools directly on the page, for example, for searching products (search_products). Instead of randomly clicking through pages, the agent calls precisely the tool it needs, making purchases faster, cheaper, and far more reliable. WebMCP is currently experimental and in an early preview phase. It is available behind experimental flags in the Chrome browser (chrome://flags/#enable-webmcp-testing), in Chrome Origin Trials, in some specialized agent browsers, and in development environments.
A beta version of WebMCP implementation at the edge has been launched. By activating this feature, the Cloudflare platform automatically injects a small script at the network edge, making these tools available without site owners having to write a single line of code. Commerce.com, a platform specializing in flexible e-commerce solutions, uses Cloudflare technology to operate nearly 100,000 domains. When the WebMCP preview became available, Commerce.com requested its mass activation for all its stores.
According to Commerce.com representatives, it is crucial for merchants to adapt to the trend of agent-driven purchases. If a buyer asks an AI agent to find a product, they want their store to be a relevant choice. The goal is to keep the customer on the merchant's site, maintain control over the payment process, and do all this without a complete e-shop rebuild. WebMCP effectively addresses these requirements. The agent gains specific tools for product searching, detail checking, assembling a shopping cart, and obtaining a payment link, thus eliminating uncertain clicking through the page. The human buyer remains on the merchant's website, and the merchant thus retains a direct relationship with the customer. Activating this feature is designed to be secure and requires no manual template adjustments. Furthermore, WebMCP functions concurrently with server-side MCP, which supports agents connecting directly to the store (e.g., for cross-website comparison), while WebMCP supports agents operating within the buyer's browser. In both cases, the same set of shopping tools is used, so merchants do not have to choose between two systems or manage them separately.
Need a hand with your website or app?
Get in touch and I will reply within 24 hours with a proposal for the next step.
Technically, this integration is based on two main principles. Firstly, WebMCP tools are executed directly within the buyer's active browser session on the e-shop's domain. This ensures that the agent sees the same prices and cart content as a human user. No login credentials are shared with third parties, and the visit occurs exclusively on the merchant's website. Secondly, Cloudflare acts as an edge delivery layer for Commerce.com's backend MCP server. It proxies tool definitions and actions without modification, allowing merchants to retain full control over the logic of their tools.
Imagine asking your AI assistant: