Anthropic proposes a plug standard for AI. Why the boring news is the important news
The Model Context Protocol, open-sourced on 25 November 2024, is an attempt to stop every AI-to-system connection being built from scratch.
On 25 November 2024 Anthropic open-sourced the Model Context Protocol, a standard for connecting AI assistants to the places data actually lives: document stores, business tools, development environments.
It is a specification, not a product. There is nothing to buy and nothing to switch on. It is also, for an SME, potentially more consequential than most model launches this year.
The problem it addresses
Today, connecting an AI tool to your systems is bespoke work every single time. Your ERP needs one integration. Your document archive needs another. Your CRM needs a third. Each is written by someone, maintained by someone, and broken by the next update.
This is why AI pilots in SMEs so often stop at the demo. The model is fine. Wiring it to the five places your company keeps its information costs more than the benefit, so the pilot stays a pilot.
A shared protocol replaces many bespoke connections with one interface implemented once on each side. The analogy people reach for is a standard plug: unglamorous, and the reason appliances work everywhere.
Why the timing matters
Standards only work if others adopt them. A protocol published by one vendor and used by nobody else is just that vendor’s API with a nicer name. It is far too early to know which way this goes.
What is already clear is the direction of the industry’s attention. The interesting problem has moved from making models cleverer to connecting them safely to real business data. That shift favours companies with orderly systems and hurts companies whose information lives in shared drives named « final_v3 ».
What to do about it
Nothing technical. One question to add to vendor conversations from now on: how does your tool connect to our existing systems, and is that connection built on an open standard or on something proprietary to you?
And one piece of preparation that pays off regardless: the value of any connection standard depends entirely on the quality of what it connects to. Tidy master data is the unglamorous investment that makes every future AI decision cheaper.