MCP (Model Context Protocol)
An open standard that lets AI applications access data, tools and business systems.
Explanation
The Model Context Protocol (MCP) is an open standard that gives AI applications a uniform way to access external data sources, tools and systems. It is often described as a “USB-C port for AI”.
How it works
An MCP server exposes the functions of a system, such as “search customers” or “create ticket”. An AI application with an MCP client connects to the server, discovers the available tools at runtime and calls them via tool calling when needed.
Example
A sales rep asks their AI assistant: “Which open quotes over 10,000 euros expire this week?” Through an MCP server for the CRM, the AI fetches the data directly and summarizes it.
Why it matters
Without a standard, every AI application has to integrate every system separately. MCP turns many one-off integrations into reusable building blocks and lowers the effort of connecting AI to existing business systems.
What is an MCP server?
MCP defines three roles. The host is the AI application you work with, such as Claude, ChatGPT, Cursor or your own agent. The MCP client lives inside the host and speaks the protocol. The MCP server is the bridge to a specific system such as a database, CRM, file store or internal API.
An MCP server can expose three kinds of capabilities: tools (actions the model can execute), resources (data and content that can be read as context) and prompts (templates for recurring tasks). Local servers run on your own machine, while remote servers are reached over HTTP.
How to build an MCP server
Ready-made MCP servers exist for many popular services, often published by the vendor itself. For your own systems, a server can be built relatively quickly with the official SDKs, including TypeScript and Python. The real effort lies less in the protocol than in designing the tools: which actions does the AI actually need, with which parameters and which permissions?
What works well: a few clearly described tools instead of a one-to-one copy of the entire API, separate read and write permissions, per-user authentication and logging of every call. Servers from unknown sources should be reviewed before use, because an MCP server acts with exactly the permissions you grant it.
Frequently asked questions
What is MCP in AI?
MCP stands for Model Context Protocol, an open standard that lets AI applications access data, tools and business systems in a uniform way.
Who is behind MCP?
Anthropic developed MCP and released it as open source in November 2024. Since December 2025, the protocol has been developed vendor-neutrally within the Agentic AI Foundation under the Linux Foundation, which Anthropic co-founded with OpenAI and others.
What is the difference between MCP and function calling?
Function calling is the ability of a model to call tools. MCP standardizes how those tools are provided, so a server built once can be used by many AI applications.
Is MCP secure?
MCP is a protocol, so security depends on the implementation. Risks include overly broad permissions, unvetted third-party servers and prompt injection through retrieved content. Least privilege, vetted servers and approval steps for critical actions reduce them significantly.