The plain-English explanation

What WebMCP actually is

Model Context Protocol (MCP) is an open protocol Anthropic released in November 2024 for connecting AI assistants to tools and data. It has since been donated to the Agentic AI Foundation under the Linux Foundation. Think of it as a universal plug for AI: a standard interface that lets an AI assistant connect to a tool or data source without custom integration work.

WebMCP is a separate proposal. It is a browser API, drafted in the W3C Web Machine Learning Community Group with editors from Microsoft and Google, that lets a web page offer JavaScript-based tools (actions) to AI agents. It is a draft community report, not a W3C standard. When an AI agent needs to find information about a business (what services it offers, where it operates, how to contact it, what it costs), it can either infer that information by reading unstructured web pages (the current default) or, where a site supports it, use the tools the site offers directly. The second path is designed to be faster and more accurate.

Your website
Designed for human readers. Layout, images, prose. AI must infer your business details by reading it like a human, slowly and imperfectly.
Google Business Profile
Structured but limited. Fixed fields: name, address, hours, category. Good foundation; doesn't cover service details, pricing, or dynamic information.
WebMCP-ready website
A machine-readable interface AI agents can use directly, where supported. It can answer structured questions (services, areas served, pricing tiers, availability) with less inference required.

A useful analogy: WebMCP may become to AI agents what Google Places listings were to local search in 2010. Many businesses were slow to set one up, and those that did often gained an advantage. WebMCP is at an even earlier stage. It is still a proposal, so for most businesses the sensible move is to get the foundations ready rather than rush.

How it works

What happens when an AI agent queries your business

→
User asks an AI agent a business question"Find me a reliable commercial electrician in Red Deer, Alberta who does industrial projects." The agent interprets the query and determines it needs to retrieve specific business data.
→
Agent queries available data sourcesWithout WebMCP: The agent browses web pages, reads unstructured content, and tries to extract relevant details. With WebMCP, where the site and browser support it: The agent uses tools the site offers, receiving structured responses that directly answer what it asked.
→
Agent synthesizes and delivers a recommendationBusinesses with clean, structured information are easier for an agent to represent accurately. No AI provider has documented using WebMCP to decide which business to recommend, but clear, structured information makes it less likely an agent gets your details wrong.
→
User receives a direct, named recommendationNot a link list, but a specific business with relevant details: what they do, where they serve, how to contact them. A business whose information is hard for the agent to read may be represented less accurately, or not at all.

The opportunity

Why it's worth preparing now

WebMCP is the newest layer of the AI-era visibility stack. SEO has been a known discipline since the early 2000s. AEO and GEO have been discussed since 2023. Chrome opened an early preview of WebMCP in February 2026 and is running an origin trial. Adoption is at its earliest stage.

For Alberta small and medium businesses, this means:

  • Most businesses haven't started on the machine-readable layer. Unlike SEO, where established businesses have years of ranking history to defend, this layer is a fairly level playing field right now. On it, a two-year-old business that does the work can be ahead of a 20-year-old business that hasn't.
  • MCP support is spreading. Claude, Cursor, GitHub Copilot and ChatGPT (OpenAI announced MCP support in March 2025) all support MCP. These tools connect to MCP servers that a user chooses to add; they don't discover businesses on their own. WebMCP, the separate browser proposal, is at an earlier stage.
  • Inference errors can hurt businesses whose information is hard to read. When AI systems must infer your business information from unstructured web content, they can get it wrong: wrong hours, incomplete service lists, outdated pricing. Clear, machine-readable information reduces that risk.
  • The setup is a defined project. Unlike SEO, which requires continuous investment in content, links, and technical maintenance, the initial setup has a clear start and end point. Because WebMCP is still a draft, expect some updates as the proposal changes.

We measured this rather than assuming it. Across our benchmark of 75 scored businesses scored between June and August 2026, 63 of 75 (84%) score below half on the machine-readable layer and 47% score below a quarter. Of the 25 sites we probed, none published an MCP description file, and there is no agreed standard location for one yet. Of 33 businesses whose components we logged individually, six publish an llms.txt file. The median business, local independent and Big Four alike, scores 25% of the available points on this pillar.

The service

Getting WebMCP set up

WebMCP setup is part of the Fix Sprint, a fixed-scope project that makes your business information easier for AI agents to read and use accurately. It is delivered remotely and is available to Canadian businesses of any size.

Fix Sprint
$3,900 CAD
Fixed price · Remote delivery · Report credited in full
  • MCP-compatible data endpoints for your business information
  • Configuration of llms.txt and an MCP description file (there is no agreed standard location for one yet)
  • Schema.org structured markup built and validated
  • Meta descriptions, titles and heading structure repaired
  • Google Business Profile built out
  • Robots and sitemap corrected for the named AI crawlers
  • About 90 days of stabilisation while the engines re-crawl
  • The $1,500 Discoverability Report credited in full: go ahead with the fixes and the diagnosis costs you nothing

Not sure WebMCP is your highest-impact gap? The Discoverability Report ($1,500, credited in full against the Fix Sprint) scores you across all four layers and against named local competitors. If WebMCP is where the gap is, the Fix Sprint closes it. If the gap is somewhere else, the report tells you that instead, before you spend anything on the wrong fix.

Common questions

WebMCP FAQ

Do I need technical staff to implement WebMCP?

No. The Fix Sprint is a fully managed service. You provide information about your business (services, areas served, pricing structure, contact details) and we handle all implementation, configuration, and testing. No technical knowledge required on your end.

Which AI platforms use MCP right now?

MCP is supported by Claude (Anthropic), Cursor, GitHub Copilot and ChatGPT (OpenAI announced MCP support in March 2025). These tools connect to MCP servers that a user chooses to add; they don't discover businesses on their own. WebMCP, the separate browser proposal, is in an early preview in Chrome, and no AI search provider has documented using it to decide recommendations.

Does WebMCP replace my website?

No. Your website remains important for human visitors, SEO, and brand expression. WebMCP works through your website: it is a proposed way for your site to offer structured actions to AI agents alongside the pages people read. Your pages serve humans; WebMCP tools would serve AI agents.

Is this the same as the Discoverability Report?

No. They are related but different. The Discoverability Report ($1,500) is the diagnosis: it scores you 0–100, ranks you against named local competitors, and shows which gaps are costing you. The Fix Sprint ($3,900) is the implementation, and WebMCP is one of the gaps it closes. The Report is credited in full against the Fix Sprint, so if you go ahead with the work the diagnosis costs you nothing.