do things?
ChatGPT, Gemini, Copilot, Claude, ... we use them to brainstorm, answer questions, write code, analyse data, ... but ask them to:
🔹 Pull a report from your CRM
🔹 Perform a calculation using company data
🔹 Call a service that is relevant to your business process
...and you are back to building custom integrations.
The problem? AI models are isolated.
Every connection to the outside world requires custom work. Every tool, every data source, every workflow: each time, you have to integrate again and write custom code. Model Context Protocol (MCP) changes that.
MCP is an open standard, launched by Anthropic about a year ago, that lets AI models communicate with external sources through a uniform interface. It is therefore often described as the "USB-C port for AI": one standard, countless possibilities.
In a recent live demo, I showed how one AI host worked with two independent MCP servers (calculator + weather API) to answer more complex questions. No integration code between those servers. No custom adapters. Admittedly a very simple setup, but the sometimes challenging and surprising questions from the audience demonstrated the robustness of the system (see answers in the comments):
- "Give me the capitals of Europe starting with the letter B, and calculate their current average temperature?"
- "What is the weather like at Hogwarts?" (Harry Potter's school)
Seemingly trick questions, but AI handled these traps effortlessly to formulate a correct and well-supported answer, without extra context or additional help. You give AI a toolbox with all kinds of tools, accompanied by instructions on how to use them, and it gets to work with them effortlessly to arrive at a good answer.
The implications of this technology are enormous:
💎 For developers: build an MCP server once, reuse it everywhere.
💎 For strategists and architects: AI becomes the orchestrator of your entire ecosystem.
💎 For companies: scalable AI integration without vendor lock-in.
We are evolving from "AI as assistant" to "AI as platform", and MCP is one of the building blocks making that possible. The question is not whether this shift will happen, but how quickly. How are you preparing your organisation for composable AI?
