MCP (the Model Context Protocol) is an open standard, introduced by Anthropic in late 2024, that gives AI assistants a common way to connect to tools and data. For a business, being agent-ready means a customer's AI assistant can get accurate answers straight from you, and send requests that you confirm before anything happens.
- It's early. Most local businesses should get the basics right first.
- The simplest layer is a clear website with structured data and an llms.txt file.
- Anything that changes something, like a booking request, should wait for your okay.
What is MCP?
MCP stands for Model Context Protocol. It's an open standard that lets AI assistants connect to outside tools and data in a consistent way. Anthropic introduced it in late 2024, and support has since spread to many AI apps and developer tools.
Before MCP, every connection between an AI assistant and an outside system had to be built as a one-off. MCP gives them a shared language. People often compare it to a USB port: one standard plug that many different devices can use.
The piece that sits in front of your information is called an MCP server. It tells an assistant what it's allowed to ask, like "check service area" or "request a quote", and answers when asked. "Server" here just means a small program running on secure hosting. It isn't a box in your back office.
How is MCP different from an API?
An API is a general way for one piece of software to talk to another. Many booking, accounting and ordering systems already have one. MCP is built specifically for AI assistants: it describes what's available in a way an assistant can discover and understand on its own, and it often sits on top of an API.
Some businesses need an API so partners' software can connect. Some need MCP so AI assistants can. Some need both. The right answer depends on who's asking for your information.
What does an agent-ready business look like?
It's a business a customer's AI assistant can ask directly, and get a true answer. Customers increasingly ask assistants to find, compare, book or order for them. Picture a homeowner outside Duchesne telling one: "Find someone who can service my furnace this week and ask for a quote."
If the business is agent-ready, the assistant can:
- ask whether you serve that town
- see your services and hours
- check general availability, if you choose to share it
- send a quote or booking request with the customer's details, for you to confirm
- check the status of an order or job, for the customer who placed it
// tool call from a customer's AI assistant { "tool": "check_availability", "service": "furnace service", "town": "Duchesne" }
{
"serves_town": true,
"next_openings": ["Thu am", "Fri pm"],
"booking": "request only, owner confirms"
}
The same idea fits many kinds of business: a bakery that takes custom cake orders, a print shop answering "is my order ready?", a motel sharing which nights are open, or a parts counter confirming whether something is in stock.
How do you keep it safe?
With a propose-then-confirm model: assistants can ask questions freely, but anything that changes something waits for a person. Our own MCP servers work this way, with destructive actions left out entirely. Here is the full set of rules we recommend:
Checklist: safety rules for agent access
- Answers come from information you approved. Hours, services, service area and policies, the same facts as your website.
- Changes are proposals. A booking, quote request or order arrives as a request, and you or your staff confirm it before it becomes real.
- Destructive actions are left out. Cancelling, deleting or refunding isn't offered to assistants at all.
- Private stays private. Customer records aren't shared, and a status check only answers for the customer who placed the order.
- Everything is logged, so you can see what assistants asked and what was sent back.
- You can switch it off at any time without touching the rest of your business.
Is there a simpler first step?
Yes, and almost every business should take it first. Many assistants read the web before they answer, so make your website easy for them to read:
- clear pages for each service, with the towns you serve
- structured data (schema.org markup) with your name, phone, hours and services
- an FAQ that answers real customer questions
- an llms.txt file that sums up your business in plain words
- a complete, accurate Google Business Profile
That's the heart of GEO, and our guide to GEO walks through it. Agent access builds on top of it.
Does your business need this yet?
Probably not today, for most local businesses. It's early: support for agents is growing, and how customers use them is still settling. For nearly everyone, a good website and accurate listings matter more right now.
Checklist: signs it may be worth it sooner
- You handle a lot of requests: quotes, bookings or orders, every day.
- Customers call just to ask "is it ready?" or "do you have it in stock?"
- Your availability, inventory or prices change often, and listings go stale.
- You sell to other businesses that run their own software or AI tools.
- Your basics are already solid: website, structured data and Google Business Profile.
If none of these sound like you, do the basics and check back in a year. If several do, it's worth a conversation. Either way, we'll tell you plainly whether it's worth it for your business yet.