Agent-Ready Websites & WebMCP

A growing share of the software your customers use (their browser included) now carries an assistant that can act for them. “Book me a massage Thursday afternoon” is becoming an instruction to an agent, not a search query. That agent lands on a website and, today, reads it the way a stranger reads a paper map: screenshots, guesses, and misclicks. WebMCP is the emerging standard that lets your website stop being guessed at.

What WebMCP is

WebMCP (Web Model Context Protocol) is a proposed web standard, drafted in a W3C community group and edited by engineers from Google and Microsoft, that lets a website declare a set of named tools an AI agent may call. Instead of an agent hunting the page for a booking button, the site says: here is check availability, here is request an appointment, here is what each needs to know. The browser asks the person for consent before any tool runs; the site’s own code does the work; the agent gets a structured answer instead of a screenshot. It is the browser-side sibling of MCP, the protocol we cover on AI Agents & MCP: MCP connects agents to systems on the server; WebMCP connects them to your site inside the customer’s own session.

Why it matters for a local service business

Because the first businesses agentic booking will reward are exactly the ones whose demand arrives as appointments: spas, med spas, home services, clinics. When a customer’s assistant can complete a booking at one business and only describe a phone number at another, the structured business wins without the customer ever comparing them. Screen-reading agents also break whenever a site’s design changes; declared tools do not. This is the same argument as accessibility and mobile before it: the sites that answered the new client clearly took the demand.

Where WebMCP honestly stands

We date our status statements: as of August 2026, WebMCP is a community-group draft, not a finished standard. Chrome ships it in a public origin trial, and the main agent calling site tools today is Chrome’s own assistant; other browsers are engaged but uncommitted, and most AI agents still read screens. In other words: early. That cuts both ways: nothing about your revenue depends on it this quarter, and the cost of being ready is small while the window to be early is real. We implement it as progressive enhancement: invisible to every ordinary visitor, a few well-designed tools for the agents that ask.

What we implement

Tool design

Deciding what an agent may do on your site (check availability, request an appointment, ask a question) and, just as deliberately, what it may not. Scope is a design decision, not an accident.

Wiring to real systems

A tool is only as honest as the system behind it. We connect site tools to the platforms that hold truth (Boulevard, Mindbody, GoHighLevel) so “check availability” checks availability, not a marketing page.

Consent and guardrails

Agents read and prepare; people confirm. Payments, cancellations, and anything irreversible stay behind human review: the same judgment rule we apply to every agent we build.

Working proof

This site practices what this page says: it registers WebMCP tools of its own, so an agent that asks can read our service map and pre-fill our assessment form: and a person still presses send.

Where to go deeper

The engineering disciplines behind agent-ready work are Systems Integrations & APIs and Integrated Web Platforms. If you want your site ready before your competitors know the term, start with a systems assessment.