What is MCP? Model Context Protocol Explained

AI models are powerful — but they know nothing about what is happening in your systems right now. Model Context Protocol (MCP) is the open standard that changes that. Here is everything you need to know.

The Problem: AI Lives in a Bubble

AI models like Claude, GPT-4 and Gemini impress almost everyone who tests them. They explain complex topics, write well-formulated texts and solve logical problems with precision.

But they have one fundamental problem: they know nothing about what is actually happening in your business right now.

An AI model does not know whether an order has been placed, whether a support ticket is open, whether a customer has paid or whether a project is on schedule. It can only respond to what you feed into the conversation — not fetch real-time data, update a ticket or create a time report.

That is the difference between a tool that answers questions and a tool that actually helps you work. Model Context Protocol is the solution.

What is MCP?

Model Context Protocol (MCP) is an open standard, originally launched by Anthropic and now available as open source. The standard solves one of AI's most important infrastructure problems: how an AI model should be able to communicate with external systems and tools in a secure, structured and standardised way.

Simply put, MCP is a shared language between the AI model and your existing systems. Instead of each integration requiring a unique, custom-built solution, MCP offers a standard format that works regardless of which AI tool you use or which system you want to integrate.

It is similar to what HTTP is for the web or SFTP is for file transfers — a protocol that all parties understand and can implement once.

How MCP works – illustration

How Does MCP Work in Practice?

An MCP integration consists of three parts:

  • The MCP server — a lightweight service that sits in front of your existing system. It exposes a set of "tools" — concrete functions that the AI model can call. These might be "fetch active tickets", "create a time report" or "search among products".

  • The AI model — which in a conversation or an automated flow calls these tools when needed. If you ask the AI for "all open tickets from this week", it knows to call the right tool, fetch data in real time and present the result.

  • Your existing infrastructure — your database, your API, your business system. It remains completely unchanged. The MCP server is built on top of what you already have.

This means you do not need to change systems, rewrite code or invest in new platforms. You build a layer on top of what already works — and open the door to an entirely new way of interacting with your systems.

Smart Source's Four MCP Implementations

At Smart Source we have implemented MCP in four of our own production systems. These are not pilot projects or proof-of-concepts — they are systems used actively every day.

Eventry – Ticketing System with AI Assistant

Eventry's ticketing system received an AI assistant via MCP. We built 21 live endpoints in four weeks, without touching a single line of their existing code. The AI can now search among events, handle ticket enquiries and provide real-time answers directly from the system's data.

Umbraco CMS – AI That Reads and Writes Content

Our Umbraco MCP server enables AI models to read and write content directly in the CMS. It is used to publish articles, update pages and manage media libraries via a conversational interface — without opening the backoffice tool.

Kimai – Time Tracking via AI

With Kimai MCP, AI assistants can log time, fetch time reports and analyse project time directly. What previously required manual entry in a time tracking system now happens in a natural conversation.

Krayin CRM – AI in the Sales Process

Krayin CRM is connected via MCP, meaning AI can create leads, update contact information, manage newsletters and analyse the sales pipeline directly — without leaving the conversation.

MCP or RAG – Which Should You Choose?

RAG (Retrieval-Augmented Generation) is often mentioned in the same breath as MCP, but they solve fundamentally different problems:

  • RAG is right when you want the AI to search and answer questions based on documents, knowledge bases or PDFs. The AI reads — but does not act in your systems.

  • MCP is right when you want the AI to be able to do things in real time — create an order, update a ticket, fetch current data, log an activity.

Most organisations actually need both — RAG for knowledge management and MCP for system integration. The key is to start with the right architecture for the right need. Read our comparison article: RAG vs MCP – Which Should You Choose?

What Does MCP Mean for Your Business?

If your organisation uses AI tools today — or is planning to — MCP is the architecture that makes the difference between an AI that answers questions and an AI that actually helps you work.

You do not need to change your business system. You do not need to wait for your system vendor to launch an AI integration. You build the MCP server on top of what you already have and open the door to an entirely new way of interacting with your systems.

We have built MCP solutions for ticketing systems, CMS, time tracking and CRM. The next system could be yours.

Want to see a demo of MCP connected to one of your systems? Contact us for a free 30-minute meeting — we will show you how it works in practice.

Mattias Jacobsson, CEO Smart Source AB – Malmö
Mattias Jacobsson, CEO Smart Source AB

Frequently Asked Questions about MCP

What does MCP stand for?

MCP stands for Model Context Protocol – an open standard that allows AI models to communicate directly with external systems and tools via standardised APIs. The standard was launched by Anthropic and is now available as open source.

Do I need to change my business system to use MCP?

No. The MCP server is built on top of your existing system and does not change anything in your existing infrastructure. You add a layer that exposes your system's functions to AI models — without needing to rewrite or replace what already works.

How long does it take to build an MCP integration?

It depends on the complexity of the system, but for a well-structured API a basic MCP server can be ready in 2–4 weeks. We built 21 live endpoints for Eventry in 4 weeks without touching their existing code.

What is the difference between MCP and RAG?

RAG (Retrieval-Augmented Generation) is used when AI should search and answer questions based on documents and knowledge bases — the AI reads but does not act. MCP is used when AI should perform actions in real time, such as creating an order, updating a ticket or logging time. Most organisations need both technologies.

Is MCP safe to use in production environments?

Yes. The MCP server you build controls exactly which tools and actions the AI model has access to. You define what the AI is allowed to do — and what it is not. Security and permission management are a central part of MCP architecture.