All posts

AI

Ready for AI agents: building with MCP and UCP

Lunar StudioEngineering3 min read

AI agents are becoming a new kind of customer. How we use MCP and UCP to make products and shops ready for them, safely.

AI agents are starting to do real work for people. They compare products, book tables and fill in forms. To do that well, they need a reliable way to use your services. Two open standards now make that possible: MCP and UCP.

What MCP does for AI agents

The Model Context Protocol (MCP) is an open standard for connecting AI agents to tools and data. Instead of building a separate integration for every assistant, you build one MCP server. Any compatible agent can then discover what it offers and use it.

An MCP server describes what an agent can do with your service. For example, a booking system might offer one action to check availability and another to reserve a slot. The agent reads those descriptions and picks the right one for the task.

What UCP adds for commerce

The Universal Commerce Protocol (UCP) is newer. Google and Shopify launched it in January 2026 as an open standard for agentic commerce. It lets an agent find products, build a basket and complete a checkout on a shopper’s behalf.

A shop publishes a short profile that tells agents what it supports, such as checkout, discounts or order tracking. UCP also works over MCP, so it sits alongside an MCP server rather than replacing it.

How we connect AI agents

  1. Start with the jobs. We list the tasks a customer might hand to an agent, and the information each task needs.
  2. Expose less, not more. Every action is a door into your systems, so we only open the ones that serve a real task.
  3. Keep people in control. Payments, bookings and personal data get clear confirmation steps. Nothing happens without the customer’s say.
  4. Write for machines and people. Clear names and descriptions help agents choose the right action. They also make the integration easier for your team to maintain.
  5. Test with real agents. We try the integration with several assistants, watch where they stumble and refine it.

If your service is hard for an agent to use, the agent will simply recommend someone else.

Security comes first

Agent access is still access. So we apply the same standards we use everywhere else: every request is authenticated, permissions stay as small as possible, and activity is logged. Our ISO 27001 certification makes these checks part of how we work. We wrote more about this in Security by design.

Where to start

You don’t need to rebuild anything. Most integrations sit on top of the systems you already have. A content site can start with a read-only MCP server for its articles. A shop can add UCP to its existing checkout. Our headless WordPress builds already keep content separate from design, which makes this step much easier.

AI agents are a new kind of visitor, and they are arriving fast. If you want your product ready for them, get in touch and we’ll plan the first step together.

Next post

Teaching a machine to tell a bedtime story

AI