Technical overview

What is

MCP?

A plain-language guide to the standard behind AI with boundaries.

 Back to AI with boundaries
The basics

One standard to connect

AI to your tools.

MCP stands for Model Context Protocol. It is an open standard, introduced by Anthropic in 2024, that defines one common way for AI applications to talk to external tools and data.

 

Think of it as USB-C for AI: instead of a custom connection for every pair of AI and tool, there is one plug that fits both.

How it works

Three parts,

one conversation.

Here is what happens when your team asks an AI to do something.

The AI app

The assistant or agent your team already uses. It asks the server what it may do and requests actions on a person's behalf.

The MCP server

A small service between the AI and your systems. It decides which actions exist and what happens when they are called.

Your tools and data

Warehouse, accounting, calendar, CRM. The server connects to them; the AI never touches them directly.

The key point

The server sets

the limits.

MCP standardises the connection, not the rules. What an AI can do depends entirely on what the server offers.

 

That is why every Timelyx setup is custom-built: we only expose the actions you have agreed to.

What we build

Four layers of

control.

Each one is set up around your rules.

Scoped tools

Only the agreed actions exist. Everything else is simply not available to the AI.

Read or change

Reading data and changing data are separate permissions, set per action.

Approval gates

Important actions pause until a person approves, with limits such as order value.

Audit log

Every request and result is recorded, so you can always see what the AI did.

An example: the Saturday order

 

1. An employee asks the assistant: check stock and prepare this week's order.

2. The AI uses a read-only tool to check stock. Allowed.

3. It uses another tool to prepare a draft order. Allowed.

4. It tries to send the order. The server holds it and asks the manager to approve.

5. The manager approves in one click. The order is sent and the action is logged.

6. The AI later tries to change the supplier. That tool does not exist, so nothing happens.

Technical questions

 

Is MCP only for developers?

No. Developers build the server; your team simply uses their AI assistant as usual.

 

Does the AI see all my data?

Only what the server exposes. That scope is part of what we agree with you.

 

Can the rules change later?

Yes. Rules are adjusted as your business changes.

 

Is MCP proprietary?

No. It is an open standard that many AI applications support.

Want to see this

for your company?

Book a consultation and tell us what AI should never do.

 Book a consultation