Originally published on The Searchless Journal
The web is about to get a new layer of infrastructure, and it has nothing to do with how pages look or how fast they load. It has everything to do with how AI agents interact with those pages on your behalf.
WebMCP, a W3C Community Group Draft co-authored by engineers from Google and Microsoft, is currently shipping in Chrome 146 beta. It defines a standardized protocol for AI agents to discover and execute actions on websites, replacing the brittle DOM-scraping approaches that agents rely on today.
If that sounds abstract, consider the concrete implications: within 12 to 18 months, AI agents could book your hotel, fill your prescription, compare insurance quotes, and complete your purchase, all without ever rendering a single webpage. WebMCP is the plumbing that makes that possible.
What Is WebMCP?
WebMCP stands for Web Model Context Protocol. It extends the MCP (Model Context Protocol) concept, originally developed by Anthropic for connecting AI models to external tools, into the browser environment. The key difference: instead of an AI model calling an API endpoint, WebMCP lets agents interact directly with websites through a structured capability contract.
Think of it as the difference between asking someone to navigate a physical store (current scraping approach) and giving them a catalog with order forms (WebMCP approach). One is fragile, slow, and error-prone. The other is structured, reliable, and fast.
The W3C Community Group Draft specification defines two capability modes:
Declarative mode handles simple, form-like interactions. A website exposes a set of structured actions, such as "search products," "add to cart," or "book appointment," along with the parameters each action requires. The agent fills in the parameters and receives a structured response without ever needing to parse HTML.
Programmatic mode handles complex multi-step flows. For actions that require sequential decisions, such as selecting a flight, choosing a seat, adding baggage, and paying, the website exposes a state machine that the agent can navigate step by step.
Both modes replace the current approach where agents attempt to simulate human browsing by reading DOM elements, clicking buttons, and hoping nothing changes.
Why Google and Microsoft Are Co-Authoring This
The fact that Google and Microsoft are co-authoring WebMCP is significant. These two companies compete aggressively in search, browsers, cloud, and AI. But they agree on one thing: the current approach to agent-web interaction is broken.
Today, when an AI agent tries to interact with a website, it essentially pretends to be a human using a browser. It renders the page, reads the DOM, figures out which buttons to click, and hopes the site has not changed since last time. This approach is:
- Fragile: Any redesign breaks the agent's understanding of the page.
- Expensive: Rendering a full page to extract structured data wastes compute.
- Slow: Multiple round-trips for what should be a single API call.
- Opaque: Website owners have no visibility into or control over how agents use their site.
Google has a direct incentive to fix this. If AI agents become a primary interface for web commerce, Google needs agents to find and transact with Google Shopping merchants, Google Flights listings, and Google Ads partners. A structured protocol ensures agents can reliably complete these transactions.
Microsoft has a parallel incentive through Copilot and Bing. If Copilot is going to book your travel, order your groceries, and manage your calendar, it needs a reliable way to interact with the services behind those tasks.
The W3C Community Group process means this is not a proprietary Google or Microsoft standard. It is being developed in the open, with input from the broader web community. But the fact that the two largest browser vendors are aligned gives it an unusually high probability of becoming a real standard.
Chrome 146 Beta: The First Implementation
Chrome 146 beta shipped the first browser-level WebMCP implementation in early May 2026. This is not a full production rollout, but it is a functional preview that developers can test.
What Chrome 146 includes:
- Capability discovery: Websites can declare their WebMCP capabilities through a well-known path, similar to how robots.txt declares crawling rules or how llms.txt declares AI-readable content.
- Action execution: Agents can invoke declared actions and receive structured responses.
- Permission model: Users retain control over which actions agents can execute, with a browser-level permission prompt.
- Session management: Multi-step interactions maintain state across sequential actions.
The permission model is critical. WebMCP does not give agents unchecked access to websites. When an agent wants to execute an action, the browser presents a permission prompt to the user. Think of it like the existing camera or location permission prompts, but for "allow this agent to book a hotel on your behalf."
This user-consent layer is what separates WebMCP from unchecked scraping. Websites control what capabilities they expose. Users control which capabilities agents can use. Agents operate within the contract.
How WebMCP Differs From Existing Standards
WebMCP is not the first attempt to make the web more machine-readable. Here is how it compares to existing approaches:
Schema.org / Structured Data: Schema.org helps search engines understand what a page contains (a product, a recipe, an event). WebMCP helps agents understand what a page can do (search, book, purchase). Schema.org is about content description. WebMCP is about action execution.
llms.txt: The llms.txt convention gives AI models a readable summary of a website's content. It is a static file that tells an AI "here is what this site is about." WebMCP is a dynamic protocol that lets an AI actually interact with the site. llms.txt is a brochure. WebMCP is a front desk.
OpenAPI / REST APIs: Traditional APIs require developers to write custom integrations. Each API has its own authentication, data formats, and error handling. WebMCP standardizes the interaction layer so agents can work with any WebMCP-enabled site without custom integration for each one.
Stripe Agentic Commerce Protocol: Stripe's protocol, announced recently, focuses specifically on payment infrastructure for AI agents: how an agent securely pays for something on your behalf. WebMCP operates at a different layer. It handles the entire interaction (discovery, selection, configuration) leading up to the payment. The two protocols are complementary, not competing.
Google Discovery v5: Discovery v5 is Google's internal framework for how AI agents discover and interact with content. WebMCP appears to be the public, standardized implementation of some of these concepts, moved into the W3C process for broader adoption.
What This Means for Ecommerce Brands
If you sell anything online, WebMCP has direct implications for how customers will find and buy your products within the next 12 to 24 months.
Today, a customer asks an AI assistant "find me a camping stove under $200 with good reviews." The agent searches the web, scrapes product pages from multiple retailers, tries to extract prices and reviews, and presents a comparison. The process is slow, unreliable, and prone to errors from page redesigns or missing data.
With WebMCP, the same query works differently:
- The agent discovers which camping retailers expose WebMCP capabilities.
- It sends a structured "search products" request to each retailer with parameters: category=camping stoves, max_price=200, sort=rating.
- Each retailer returns structured product data: name, price, rating, availability, image URL.
- The agent presents a clean comparison and, with user permission, can proceed to "add to cart" and "checkout" actions on the chosen retailer.
For the retailer, this means:
- Your product catalog becomes directly queryable by AI agents without relying on them correctly parsing your HTML.
- Your checkout process becomes an API call rather than a fragile sequence of form fills and button clicks.
- Your brand appears in agent-driven commerce alongside (or instead of) traditional search results.
The brands that implement WebMCP early will have a structural advantage in agent-driven commerce, just as brands that implemented structured data early had an advantage in rich snippets.
What You Should Do Now
WebMCP is in beta. Full production rollout is months away. But there are concrete steps you can take today to prepare:
1. Audit your transactional flows. List every action a customer can take on your website: search, filter, compare, add to cart, checkout, book, schedule, subscribe. Each of these will eventually become a WebMCP capability.
2. Document your capability schema. For each transactional flow, define the inputs (what parameters does the action require?) and outputs (what structured data does it return?). This documentation becomes the basis for your WebMCP capability manifest.
3. Prepare a skills file. Similar to how llms.txt describes your site's content for AI models, a skills file will describe your site's capabilities for WebMCP agents. Start drafting this now.
4. Test with Chrome 146. Install Chrome 146 beta and experiment with the WebMCP preview. The W3C Community Group has published example implementations that show how to declare capabilities and handle agent requests.
5. Monitor the W3C draft. The specification will evolve as it moves through the standardization process. Subscribe to the WebMCP Community Group mailing list and track changes to the draft.
The Strategic Landscape
WebMCP does not exist in isolation. It is part of a broader shift toward the agentic web, where AI agents become primary intermediaries between users and online services.
Three infrastructure layers are emerging:
- Discovery layer: How agents find relevant services (traditional search, AI search, agent directories).
- Interaction layer: How agents interact with those services (WebMCP).
- Payment layer: How agents pay for transactions (Stripe Agentic Commerce Protocol, traditional payment APIs).
Brands that optimize for all three layers will be positioned to capture the growing share of commerce driven by AI agents. Brands that optimize only for traditional search will find themselves increasingly invisible in agent-driven transactions.
The data supports the urgency. AI-referred traffic to US retailers is up 393% year over year according to Adobe Analytics. Shopify reports AI search orders up 13x year over year. Walmart found that AI-recommended orders have 35% higher spend. The demand side is already here. The supply side, the infrastructure that lets agents reliably transact, is what WebMCP addresses.
The Timeline
Based on the current W3C process and Chrome's release cadence:
- Now (May 2026): Chrome 146 beta with WebMCP preview. Specification in Community Group Draft stage.
- Q3-Q4 2026: Likely broader Chrome implementation, potentially in stable releases. Continued specification refinement.
- Early 2027: Other browser vendors (Edge, potentially Safari, Firefox) begin implementations. Specification moves toward W3C Recommendation status.
- Mid 2027: Early-adopter brands see meaningful agent-driven commerce volume through WebMCP.
- 2028+: WebMCP becomes standard web infrastructure, expected by agents as a matter of course.
This timeline means that the brands starting preparation now will have months of lead time over competitors who wait for the standard to be finalized.
Bottom Line
WebMCP is the most significant web infrastructure development since the mobile-responsive design mandate. It changes how AI agents interact with websites from fragile scraping to structured capability contracts. Google and Microsoft co-authoring it gives it unusual credibility and adoption momentum.
Every brand with a transactional website should begin preparing now. Audit your flows, document your capabilities, draft your skills file, and test with Chrome 146 beta. The agentic web is not a future concept. The first building blocks are already shipping.
Free tools show you what is happening with your AI visibility. Searchless shows you why it is happening and what to do about it. Run a free audit to see how your brand appears across ChatGPT, Google AI Overviews, Perplexity, and more.
Top comments (0)