article — General

WebMCP: Let AI Agents Pay You Money

About this Resource

WebMCP makes websites ready for AI agents, allowing them to interact with products, services, and online tools. Explore how this emerging technology could transform the web and create new business opportunities.<h2> </h2><h2>WebMCP: Let AI Agents Pay You Money</h2>AI agents are quickly becoming a new kind of internet user. Instead of people manually browsing websites, searching for products, comparing options, filling out forms, and making purchases, AI agents could increasingly handle these tasks for them. The problem is that most websites were designed for humans, not AI agents. This is where WebMCP comes in. WebMCP is an experimental approach that allows websites to expose specific actions that AI agents can understand and use directly through the browser. Instead of an agent trying to figure out where to click by reading an entire webpage, the website can provide clear tools such as search products, compare items, add to cart, or apply a coupon. <h2>What Is WebMCP?</h2>In simple terms, WebMCP makes websites both AI-readable and AI-actionable. Traditional AI browser agents may need to inspect a website's code, analyze its page structure, take screenshots, find buttons, and simulate human clicks. This can be slow and fragile. With WebMCP, the website can provide the agent with a concise list of actions it is allowed to perform. The agent can then use those tools instead of trying to navigate the website exactly like a human. Think of it as giving an AI agent a menu of things it can actually do on a website.<h2> </h2><h2>A Real Example: AI Shopping</h2>The video demonstrates WebMCP using an espresso equipment store. Instead of manually researching different espresso machines, the user can tell their AI agent what they need. For example, they might want a machine that can make multiple drinks per day while fitting within a specific counter width. The agent can then:<ul><li>Search for suitable machines </li><li>Compare different products </li><li>Check technical specifications </li><li>Determine whether products fit the user's requirements </li><li>Find compatible accessories </li><li>Add products to the shopping cart </li><li>Apply available coupons</li></ul> The important part is that the agent isn't simply telling the user what to buy. It can actually interact with the website and perform actions while the user can see what is happening. This becomes particularly useful for complicated purchases where compatibility matters, such as computer hardware, cameras, development boards, automotive parts, or specialist equipment. <h2>Why WebMCP Is Different</h2>There are already several ways AI agents can interact with software. An agent can use a raw API, a traditional MCP server, computer-use tools, or browser automation. However, each approach comes with trade-offs. Computer-use agents may need screenshots and simulated clicks. Browser-based approaches can require the agent to inspect large amounts of website code to figure out what it can do. WebMCP sits somewhere between these approaches. It allows a website to retain its normal visual interface while also providing structured tools that external AI agents can use. This creates an interesting model, the website keeps its UI for humans, while also becoming usable by the user's own AI agent. <h2>Your Agent, Not the Company's Agent</h2>One of the biggest ideas behind WebMCP is the concept of bringing your own agent. Instead of every website forcing users to interact with its own built-in AI assistant, WebMCP allows users to bring the AI agent they already use. That could eventually mean a person has one main AI agent that understands their preferences, context, and history, and that agent can interact with many different websites. The website simply provides the capabilities. This could be especially important as AI agents become more common in everyday work and personal tasks. <h2>WebMCP Can Use Your Existing Browser Session</h2>Another interesting feature is that WebMCP can make tools conditional based on the browser session. For example, a website could provide certain tools to everyone, while additional tools become available after the user logs in. Because the agent operates through the browser session, the website can use the user's existing authenticated state rather than requiring the agent to separately manage API keys and credentials. This could make agent access easier for websites that don't want to build an entirely separate API infrastructure. <h2>Where Could WebMCP Be Useful?</h2>The possibilities go well beyond online shopping. E-commerce - Web stores could allow agents to search products, compare specifications, check compatibility, manage carts, and complete other shopping tasks. SaaS and Business Software - Dashboards, analytics platforms, marketing software, and administrative systems could expose useful actions to agents, making repetitive business processes easier to automate. Regulated Services - The approach could also potentially be used for more controlled applications, such as insurance or banking self-service. For example, websites could expose read-only tools that allow an agent to retrieve information without giving it unrestricted control. Internal Company Tools - Businesses could also use WebMCP internally, allowing employees' agents to interact with company systems and automate repetitive workflows. <h2>The Business Opportunity</h2>The bigger opportunity may not simply be building WebMCP technology. It could be helping existing businesses become ready for AI agents. One idea discussed in the video is an agent conversion agency. Instead of selling businesses a completely new website, the agency could help existing businesses make important customer journeys accessible to AI agents. For example, a law firm might allow an agent to request a consultation, while a home-services company could let an agent request a quote or schedule a service. The agency could charge an initial setup fee and then provide ongoing monitoring and improvements. Another opportunity is an AI agent mystery shopper. The idea is simple, use AI agents to test whether important tasks on a website actually work. An agent could attempt to purchase a product, book an appointment, request an insurance quote, or complete another important customer journey. The business would then receive a report showing where the agent got stuck, what information was missing, and what could prevent an AI-driven customer from completing the task. <h2>WebMCP Is Still Early</h2>There is an important catch, WebMCP is experimental technology. The feature is still in the early stages, with support being developed around Chrome and involving Google and Microsoft. The example shown in the video can be experimented with, and developers can also use the associated code repository to build their own versions. That means WebMCP isn't yet something every business should immediately rebuild its entire website around. But its early stage is also what makes it interesting.<h2> </h2><h2>The Bigger Picture</h2>The internet has traditionally been optimized around humans. Search engines needed to understand webpages. Then AI systems began answering questions about those webpages. The next step could be making websites actionable by AI agents. The question is no longer just; can AI understand my website? It becomes, can an AI agent actually complete a task on my website? That shift could have major implications for e-commerce, SaaS, customer service, business automation, and online transactions. As more people begin relying on personal AI agents, businesses may need to think of agents as a new category of internet visitor and make sure those visitors can actually do something useful. <h2>Key Takeaways</h2><ul><li>WebMCP makes websites easier for AI agents to understand and interact with.</li><li>Instead of relying entirely on screenshots, DOM inspection, and simulated clicks, agents can use structured website tools. </li><li>Users can potentially bring their own AI agent rather than relying on an agent built into each website. </li><li>WebMCP could be particularly useful for complex shopping, SaaS, business workflows, and internal tools. </li><li>Businesses could create new services around making websites agent-ready and testing whether AI agents can successfully complete important tasks. </li><li>The technology is still experimental and early, but that also creates an opportunity for developers and entrepreneurs willing to experiment now.</li></ul> <h2>Final Thoughts</h2>WebMCP points toward a potentially important change in how people interact with the internet. Today, we visit websites and operate them ourselves. In the future, we may increasingly tell our AI agents what we want and let them handle the interaction. For businesses, that creates a new challenge, don't just build websites that humans can use, build websites that AI agents can successfully work with too. And for developers and entrepreneurs, that could create an entirely new category of opportunities around the agent-native web.