TECHNOLOGY / EXPLAINED

How Commerce.com and Cloudflare are Prepared Storefronts for AI Agents Using WebMCP

2026-10-27 · GlobalTech Daily Editorial Desk · Source: blog.cloudflare.com

Original editorial illustration for How Commerce.com and Cloudflare are Prepared Storefronts for AI Agents Using WebMCP
Key takeaways Factual Overview and Introduction According to an official blog post by Jenny Lea, Max Brosnahan, and Matthew Conroy, websites were built for humans rather than AI agents. Most shopping agents currently operate using a site's HTML, which makes the process slow, token-

Factual Overview and Introduction

According to an official blog post by Jenny Lea, Max Brosnahan, and Matthew Conroy, websites were built for humans rather than AI agents. Most shopping agents currently operate using a site's HTML, which makes the process slow, token-heavy, and fragile against theme changes. To address this, Google introduced WebMCP, an experimental browser standard providing structured tools on a page, such as search_products. WebMCP is currently available behind experimental Chrome flags, Chrome Origin Trials, select agentic browsers, and developer environments like Cloudflare Kitesurf. In August, Cloudflare launched a beta edge implementation of WebMCP that automatically injects a tiny script at the edge without requiring site owners to write code.

Partnership with Commerce.com

Commerce.com, a commerce platform running nearly 100,000 domains on Cloudflare, partnered to enable WebMCP across their stores. Kristina Pototska, Product Management Director at Commerce.com, noted that merchants want their stores included when shoppers ask agents to find products, while retaining customers, checkout control, and avoiding site rebuilds. WebMCP allows agents to use real tools for searching products, checking details, building carts, and obtaining checkout links. Furthermore, WebMCP works alongside server-side MCP, covering both browser-based agents and direct-connection assistants using the same set of shopping tools.

Technical Integration and Execution

The integration relies on two primary principles. First, WebMCP tools execute directly inside the shopper's active browser session on the storefront's domain, ensuring identical prices and cart visibility without sharing credentials with third parties. Second, Cloudflare serves as the edge delivery layer for Commerce.com's backend MCP server, proxying tool definitions and actions unchanged. For example, instead of simulating clicks and facing bot challenges, an agent can detect store tools via document.modelContext and make direct calls like search_products and add_to_cart. The user completes the final checkout manually to keep payment details secure.

Scaling Challenges and Custom Opt-Outs

Enabling WebMCP across nearly 100,000 storefronts presented a challenge because WebMCP was initially designed as a single on-or-off switch for an entire domain zone. Commerce.com wanted to use an edge bridge toolpack to proxy tool requests between their backend MCP server and the visitor's browser while allowing individual merchants to opt out. Cloudflare solved this by utilizing Configuration Rules and custom hostname metadata. If a merchant opts out in their dashboard, Cloudflare bypasses WebMCP for that specific store at the edge, resulting in zero extra latency and no injected scripts.

Custom Endpoints, Fallbacks, and Future Outlook

Real-world implementation led to additional practical refinements. Since Commerce.com's server resides at /api/mcp rather than the default /mcp path, Cloudflare added a configurable MCP URL setting at the zone level. Additionally, seamless fallbacks were implemented so that if a store's MCP server goes offline or a browser lacks WebMCP support, the script exits without disrupting normal storefront loading. Cloudflare and Commerce.com plan to continue investing in WebMCP, and representatives are scheduled to speak at Cloudflare Connect in San Francisco on October 21, 2026.

Read the original source: How Commerce.com and Cloudflare are getting storefronts ready for AI agents

AI-assisted reporting; source claims are attributed and have not been independently verified.

More to read