What is MCP? Model Context Protocol Explained Simply

What is MCP

MCP stands for Model Context Protocol. It is an open standard that defines how AI models connect to external tools and data sources - your accounting software, databases, file systems, project boards. Anthropic released it in late 2024, and it has since been adopted across the industry, so the same connector works with different AI assistants.

The easiest analogy is USB-C for AI. Before USB-C, every device needed its own cable. Before MCP, every AI-to-tool integration was a custom build. With MCP, a tool exposes one standard connector - an MCP server - and any MCP-capable AI assistant can plug into it. Xero, GitHub, Notion, and hundreds of other products now have MCP servers.

How MCP works

There are three parts: the AI assistant (the client), an MCP server for each tool, and the protocol that connects them.

1. An MCP server sits in front of a tool - say your accounting system - and advertises what it can do: list invoices, create a draft invoice, fetch a contact.
2. You connect that server to your AI assistant and authorise it, typically with the same login and permissions you already have.
3. When you ask the AI something like "which invoices are overdue", the model sees the available actions, calls the right one, and gets structured data back.
4. The AI works with the result - summarising, cross-referencing another tool, or drafting the follow-up emails - all in one conversation.

The practical effect is that the AI stops being a text box you paste things into and becomes something that works inside your actual systems. One conversation can span your inbox, your accounts, and your database because each is just another MCP connection.

What MCP means for NZ small businesses

For a small business, MCP is the difference between "AI that gives advice" and "AI that does the work". The copy-paste tax disappears: no more exporting a report, pasting it into a chat, copying the answer back out.

Concrete examples. A business owner could connect Claude to Xero via MCP and ask "who has not paid this month" then have it draft the reminder emails - a monthly admin session shrinking to a short conversation. A trades or services firm could connect its job management and calendar tools so the AI can summarise open jobs and schedule follow-ups. A small development or marketing team could wire the AI into its database and project board so reporting queries get answered in plain English instead of waiting on whoever knows SQL.

Because MCP is an open standard, you are not locked in. The connectors you set up today keep working as AI assistants improve, and switching assistants does not mean rebuilding your integrations.

Getting started

The fastest path is to use connectors that already exist rather than building your own. Most major business tools now ship an official MCP server or connector, and assistants like Claude support them out of the box.

• List the two or three tools where you spend most of your admin time - accounting, email, job management.
• Check whether each has an official MCP server or connector - most major platforms do in 2026.
• Connect one read-only workflow first, like asking questions of your accounting data, before allowing any write actions.
• Keep the human-approves rule for anything that sends, pays, or deletes.

Frequently asked questions

Related terms

Related articles

• Stop Copying and Pasting: How MCP Turned My 1-Hour Admin Into a 3-Minute Conversation
• How to Automate Your Xero Accounting with Claude AI
• How to Connect Cursor and Supabase Using MCP