In Simple Terms: What is MCP protocol?
Imagine you have an assistant that helps you with everyday work. You’d like to give the assistant its first task: filtering your email‑subscriber database according to specific criteria. To do that, the assistant needs access to the subscribers list. One option is to call your email‑marketing specialist, ask for the list, and then forward it to the assistant. A quicker option is to share the specialist’s contact details with the assistant so it can request the list directly – saving you time.
With each new assignment, the assistant will need more and more knowledge about your specialists. For instance, if you want a list of your top customers, the assistant must reach out to your sales manager. To simplify this, you create a short form and ask every specialist to fill it in:
- Role
- Contact information
- How they can help (the data they can supply or the actions they can perform)
After all the forms are filled out, the assistant now has a directory of specialists:
| Specialist | Contact method | How they can help |
|---|---|---|
| Sales manager | Phone | • Launch any product for sale • Provide information about customers and orders |
| Email‑marketing specialist | • Send an email newsletter • Provide information about email subscribers |
Now suppose you set a new task: Send an upsell email to customers whose total order amount is less than $50. The assistant checks the directory and sees that:
- The sales manager can provide customer and order data, and
- The email‑marketing specialist can send newsletters.
The assistant asks the sales manager for the filtered customer list, then forwards it to the email‑marketing specialist with instructions to send the upsell campaign. Because of the directory, the assistant works autonomously – choosing the right specialist for each request without further input from you.
And now imagine that your assistant is some AI tool that you use. You want it to work with your data or perform certain actions with third-party systems. So, the MCP protocol is like the form filled out by your specialists – it helps the AI tool understand what data it can access and what actions it can perform.
Suppose you want the AI tool to access your customer database. You would then create an MCP server (this acts as your specialist), describe in a specific format that the MCP server can provide access to the customer database, and specify which command the AI tool can use to request that data. If the AI tool supports MCP integration, you simply connect your MCP server to it. Then, when you’re working with the AI tool, you can ask it to find any customer in your database – the AI tool will call the appropriate command on the MCP server, receive the customer data, and show you the result.
You can connect many MCP servers to a single AI tool. Likewise, the same MCP server can be connected to multiple AI tools.
Why do we even need the MCP protocol? To provide some data or allow certain actions by integrating MCP servers into an AI tool – it’s all about improving your experience with the AI. The MCP protocol itself is designed to give AI programs an easy way to integrate with your own data or applications. Think of it like giving ChatGPT or any AI tool the ability to “plug in” to your calendar, your internal CRM, or a specific knowledge base.