What is an MCP server, and why your business needs one
MCP is the piece that lets an AI assistant like Claude look into your systems. Here is what it is, without jargon, and what to know before you install one.
By Jean Lussier · 2 min read
The problem MCP solves
An AI assistant knows nothing about your business. To answer a question about your orders, clients or inventory, it has to be able to look into your systems. Until recently, every connection between an AI and a piece of software had to be custom-built, a different way each time.
What MCP is
The Model Context Protocol (MCP) is an open standard, introduced by Anthropic in November 2024, that defines a common way to connect an AI assistant to tools and data. It is often compared to a USB-C port for AI: one type of plug, for a host of devices.
There are two sides:
- The MCP client: the AI application your employees use, such as Claude.
- The MCP server: a small program that gives access to a specific system (a database, a CRM, a folder of documents) and decides exactly what the AI can see and do.
Why a business needs one
- Your data stays with you. The AI queries the system when it needs to, instead of you copying your data into an outside tool.
- You control access. The server only exposes what you choose: for example, looking up orders without being able to change them.
- Your employees ask questions in plain language. No more commissioning a report every time a new question comes up.
A concrete example
A manufacturing company in Saint-Jean-sur-Richelieu had all its information in a custom SQL database, but its employees could not query it themselves. We built an MCP server that links their AI assistant (Claude) to that database. They ask their questions in plain language, and the assistant produces the report on the spot.
Ready-made or custom server?
There are ready-made MCP servers for well-known services. They work when your needs are standard. A custom server becomes necessary when you need to connect an in-house system, apply your own access rules, or combine several data sources.
Security precautions
An MCP server opens a door between the AI and your systems. It has to be built with care:
- Least privilege: the server uses an account with only the rights it needs, read-only whenever possible.
- Sensitive data excluded: what the AI doesn't need to see is not exposed.
- Traceability: a record is kept of what is looked up.
- Where processing happens: for personal information, check where the data is processed, which Québec's Law 25 regulates.