When an AI assistant books a service for you

AI assistants such as ChatGPT and Claude can now do much of the legwork of booking a service: finding open times, comparing options, checking prices and filling in the details. The safe version of this leaves two steps with you, confirming the booking and paying for it. Before you let an assistant book anything, it helps to know how it is doing it and what to check.

This guide explains what "booking for you" means in practice, what the Model Context Protocol (MCP) is in plain words, the checks worth making, and what Singapore's own guidance on AI agents says about keeping people in charge.

What "booking for you" actually means

When people say an AI agent booked something, it usually did one of three things, or a mix of them.

1. It used the website like a person. Some assistants can open web pages, click buttons and fill in forms. This works with almost any business, but websites are built for people. Steps such as a time held for a few minutes, or a code sent to your phone, can trip an assistant up, and it may misread a page.
2. It used a connection the business provides. Some businesses offer a direct way for assistants to ask questions and take actions, such as an MCP server or an API. The assistant gets answers straight from the business's own system rather than reading pages.
3. It handed the booking back to you. Often the last step is a link with the details filled in, or a confirmation message sent to your phone, which you check and approve.

The third step is the one to look for. An arrangement where the assistant does the searching and you do the committing is both convenient and safe.

MCP in plain words

The Model Context Protocol is an open standard for connecting AI applications to outside systems. Its own site compares it to a USB-C port: one standard plug that lets an assistant connect to many different services in the same way.

For a household, the practical effect is that an assistant connected to a business's MCP server can ask that business directly: which times are open, what this visit costs, whether it covers your postal code. It no longer has to guess from a page that may be out of date.

It also means the business decides what an assistant can and cannot do. A server can offer only lookups, or lookups plus a small number of carefully limited actions. That is worth knowing, because the safest servers are the ones where nothing that costs you money happens without you.

What to watch for

Is it really the business?

An assistant can land on a lookalike page as easily as a person can. Check that the link, payment page or phone number belongs to the business itself, ideally found from its own website. Our guide to home cleaning scams covers the common tricks.

Where did the price come from?

A price read off an old page or a third-party listing may be wrong. A price from the business's own system is more reliable, but it is still worth checking that the final confirmation shows the same figure the assistant quoted.

Did it understand the day?

"Next Friday" means different things to different people, and assistants can get it wrong too. Before approving, check the date and the weekday together, plus the time and the address, including the unit number.

What does it need from you?

To book a home service, an assistant needs your address, a contact number and the details of the visit. It should never need your card number, a password, or a one-time password. Pay on the business's own payment page, reached by a link.

What can it do without asking?

Prefer services where nothing is booked or paid until you personally approve it. If a service would let an assistant commit you on its own, think twice.

What happens afterwards?

Changing or cancelling usually happens through the business's own channel, and its normal terms apply exactly as if you had booked yourself. Read them before you approve.

What Singapore's framework says

In January 2026, the Infocomm Media Development Authority launched a Model AI Governance Framework for Agentic AI. It is written for organisations that deploy AI agents, not for households, but it describes what responsible use looks like. It sets out four areas:

• assess and bound the risks upfront, for example by limiting what an agent can do and what data it can reach
• make humans meaningfully accountable, for example by triggering human approvals at significant checkpoints
• put technical controls and processes in place, including testing before an agent goes live
• enable end-user responsibility, including telling users when and how agents are used

For you as a customer, two of those translate directly. You should be told when an agent is involved, and the steps that matter, such as committing to a booking or paying, should wait for a person's approval: yours.

How HomeMaster handles bookings from AI assistants

We run an MCP server at homemaster.co/mcp that any assistant supporting remote MCP servers can add, with no key or sign-in. A connected assistant can check our prices and open times, get the exact quote, create a checkout link with the visit filled in, or send you Helen's booking card on WhatsApp.

Nothing is booked until you tap "Yes, book it" on that card, within 60 minutes, and nothing is paid until you pay by the link. Changes and cancellations happen with Helen, our AI assistant on WhatsApp, under our usual policy.

If you use ChatGPT or Claude and want to try it, our guide to booking a cleaner with ChatGPT or Claude walks through the set-up. Developers will find the full tool list in our MCP server docs, and our page for AI assistants and developers lists every route. For where AI stops and people take over, see what AI can and cannot do in home services.