Be a tool agents can call

There is a difference between an assistant mentioning your business and an assistant using it.

Mentioning is the first pillar. Someone asks for a recommendation, your name comes up, they go and find you. Using is different: the assistant checks your availability, pulls a live price, starts a booking — inside the conversation, without the person ever visiting your website.

That second thing needs you to be callable, and there is now a standard for it.

MCP, in one paragraph

The Model Context Protocol standardises how an AI assistant connects to external tools and data. It was created by Anthropic and donated to the Linux Foundation's Agentic AI Foundation in December 2025. By February 2026 it had crossed 97 million monthly SDK downloads and been adopted by Anthropic, OpenAI, Google, Microsoft and Amazon.

The practical consequence is the bit worth remembering: a server you build once works across every major assistant. That is unusual, and it is why this is worth understanding rather than waiting out.

What "being a tool" looks like for a real business

Not abstract. Concrete, by trade:

• A clinic exposes appointment availability, so an assistant can answer "can I get in Thursday?" with a real answer rather than a phone number.
• A supplier exposes live stock and trade pricing, so a builder's assistant can compare without three phone calls.
• A trades business exposes a quoting endpoint, so a rough price comes back in the conversation.
• A retailer exposes size and colour availability per store.

In each case the thing being published is not content. It is a capability, with a name, defined inputs, and a defined response.

Why this is different from having an API

You probably already have one, or your platform does. An API is written for a developer who reads documentation and writes code against it. A tool definition is written to be understood by a model at the moment it needs it — a plain-language description of what the tool does, what it needs, and what it returns.

The distance between the two is often small. If you have an API, being callable is frequently a wrapper rather than a rebuild.

The honest sequencing

Know what your platform is doing. Shopify, booking systems and practice-management software are adding MCP support now. When yours does, being callable becomes a settings toggle rather than a project — but only if you know to look for it. Ask your vendor; most have a roadmap page and almost nobody reads it.

Build it yourself if you have a live-data advantage competitors cannot match, your customers habitually ask questions only your system can answer, or you already see agent traffic in your logs. That last one is checkable today — filter your server logs by user-agent and see who is already knocking.

Get ready regardless — and this part costs nothing extra, because it is the same work as the first two pillars:

• Real prices, availability and service areas as text and structured data
• One unambiguous business entity
• Content that works without JavaScript

Machine-readable is machine-readable. Every hour spent on the first pillar buys you a head start here.

What is coming

Guides on MCP without code, wrapping an existing API as a tool, what to ask your platform vendor, and how to tell whether agents are already trying to use your website.

Tell us what would help.