Definition:
MCP (Model Context Protocol) is an open protocol that connects artificial intelligence applications to external tools and data sources. It provides a common way to exchange information and request actions, such as querying a database, reading a document or accessing a service.
An assistant can use these connections to work with information that was not part of the model’s training. MCP is neither an artificial intelligence model nor a marketing methodology: it is the protocol that enables this communication.
How MCP works
The connection runs from an AI application to a program that provides access to specific data or functions. Three components are involved:
- Host application: the AI assistant or application that the person uses and that coordinates the connections.
- MCP client: the component within that application that communicates with an MCP server.
- MCP server: the program that makes tools or information available to the application. It can run locally or on a remote system.
For example, a server might offer one tool for checking sales and another for finding products. The application identifies the available functions and can use them to handle a request, within the configured permissions and controls.
The official MCP documentation explains the protocol and its integration options.
An example of MCP in digital analytics
In a digital analytics project, a marketing team could connect an assistant to its online store’s data through an MCP server. When asked to compare sales by category across two months, the assistant would use a query tool and produce an answer based on the results.
For this to work, the integration must support that query and have authorised access to the data. If it only has read permissions, it should not allow changes to orders or prices.
This example illustrates one use of MCP in an AI agent: providing tools to carry out a specific task. The quality of the analysis will still depend on the data, how the query is formulated and how the results are interpreted.
The difference between MCP and an API
An API allows programs to communicate through defined operations. MCP provides a common structure for AI applications to discover and use tools, information resources and prompt templates.
The two can work together. An MCP server can query an analytics platform’s API and make that function available to the assistant. MCP therefore does not necessarily replace existing APIs: it can serve as an integration layer on top of them.
Limitations and permissions
Connecting an MCP server does not automatically grant access to all of an organisation’s data or guarantee correct answers. Each integration must define what information it can access and what actions it can perform.
It is also worth checking who maintains the server and what data is sent to external services. Operations that modify information, such as updating a record or sending a message, require appropriate controls and, where applicable, user confirmation.