We have a lot of conversations about connecting legacy and internal systems to LLMs thankfully it’s absolutely possible using Model Context Protocol (MCP).
Most people understand there is huge value in being able to interrogate, summarise and visualise information via an AI chat interface. LLMs give us access to “all of human knowledge” but what about your business information?
Imagine if you could leverage the power of LLMs on your data, files and systems. Model Context Protocol, or MPC for short, is the standard that enables this.
What is MCP?
MCP (Model Context Protocol) is an open-source standard for connecting AI applications to external systems. Anthropic first launched the MCP standard in November 2024 with the express aim of reducing developer reliance on building custom connectors for each tool or data source they wanted to integrate with an LLM.
The premise is that by using the MCP standard, AI applications like Claude or ChatGPT can connect to 3rd party data sources such as local files, databases, legacy software or even specially built tools such as custom calculators or propriety workflows. This enables LLMs to access key information and perform tasks.
The most basic explanation given is to imagine MCP as a USB-C port for AI applications. As USB-C provides a uniform or standardised way to connect electronic devices, MCP provides a standardised way to connect AI applications to external systems.
How does MCP work?
Think of MCP as a bridge between an AI assistant or Chat interface and business systems. The AI doesn’t need to know how each system works internally, it just needs to know how to ask for what it needs using MCP. This makes systems more modular, easier to extend, and much simpler to maintain over time.
Having an MCP server that exposes data and resources from your internal systems not only opens up the power of AI to your current system it also means you can connect them to multiple tools meaning you get to choose the best tool for the task at hand, but without having to build new integrations each time.
There are lots of publicly available MCP servers for tools like HubSpot, Slack and Salesforce. However, for SMEs, a custom MCP server is often required to bring their data and resources to the edge in a way that can be securely accessed and utilised by AI tools.
A Practical Example: Wizz MTB
Imagine you run a business that manufactures and sells mountain bike clothing and equipment.
Let’s say you’re using an AI assistant connected via MCP to the Wizz MTB system, which manages customers, products, opportunities, and support tickets.
- You ask:
- “Show me all open support tickets for our top customer and any related sales opportunities.”
Here’s how MCP enables this behind the scenes:
- The AI interprets your request:
- Identify the “top customer”
- Retrieve their open support tickets
- Check for related sales opportunities
- Using MCP, the AI calls the Wizz MTB MCP tool:
- It requests customer data
- It queries support tickets linked to that customer
- It fetches any associated opportunities
- The Wizz MTB system responds with structured data:
- Customer profile
- List of open tickets
- Active or pending opportunities
- The AI combines and presents the result:
- A clear summary of issues
- Relevant sales context
- Suggested next actions
The response is handled by the LLM and can be formatted as you please all initiated from a natural language text prompt. Pretty cool huh!
What’s happening under the hood?
When MCP is used with a system like Wizz MTB, the flow typically looks like this:
- Request is initiated
- A user asks a question or gives an instruction
- Intent is interpreted
- The AI determines which business data or actions are needed
- MCP tool is invoked
- The AI uses the Wizz MTB MCP tool to request specific data
- System processes the request
- Wizz MTB retrieves data from its internal MIS system
- Structured response is returned
- Data comes back in a consistent, machine-readable format
- Response is assembled
- The AI turns the data into a clear, human-friendly answer
Why MCP matters?
Using MCP with systems like Wizz MTB brings real advantages as it eliminates the need for custom integrations for every query. Meaning the AI (potentially multiple tools) can rely on a single standard interface to interact with the system.
This leads to faster development, since new capabilities can be added without rebuilding existing connections. It also improves data access, allowing AI to safely and consistently query business-critical systems which otherwise it would not have access to.
Ultimately the goal with a custom MCP server is to provide better decision-making giving users access to combined insights such as support, sales, and customer data all in one place and using the summarisation and visualisation power that the large LLMs offer.
Is MCP safe?
MCP itself is not inherently safe or unsafe. It is a standard for connecting systems and exchanging information between them. Like HTTPS or FTP, it defines a way for data to be shared or transported.
That said, using MCP securely depends on how it is implemented. The usual concerns still apply: authentication, authorisation, encryption, data governance, performance, and access controls all matter.
If you’re exploring how to make your internal systems AI-ready get in touch to arrange a call and we can get you on the right path.