What is MCP? Definition and meaning
MCP is short for Model Context Protocol. It is an open standard through which AI applications access external data, documents and functions. For language models, MCP therefore takes on the role that an API plays between two classic programs.
The protocol was presented by Anthropic at the end of 2024 and subsequently published as an open standard.
The structure of MCP
MCP follows a client-server principle with three roles:
- Host (= the AI application in which the language model runs)
- Client (= the connecting layer inside the host, each one linking to a single server)
- Server (= the program that makes a data source or a system available)
What an MCP server provides
An MCP server publishes three kinds of building block:
- Tools (= functions the model can run, for example creating a record)
- Resources (= data the model can read, for example files or database contents)
- Prompts (= prepared templates for recurring tasks)
Benefits over classic interfaces
Without a common standard, every combination of AI application and specialist system needs an interface of its own. MCP turns that relationship around: a system is connected once and can then be used by every MCP-capable AI application.
For business software, that means projects, times, documents or customer data can be made accessible to an assistance system without duplicating the data.
MCP vs. other meanings: a distinction
The abbreviation MCP is ambiguous. In IT training it stands for the certification title Microsoft Certified Professional, in medicine for the active substance metoclopramide.
In the software and AI field, MCP today means the Model Context Protocol and nothing else. Only this meaning is covered in this glossary.