MCPs reshape enterprise AI
Enterprise AI is moving from flashy models to a plumbing layer called “MCPs” that mediates between models and business systems to give more control in regulated environments. This architecture promises easier, safer integration for systems that must talk to HVAC, access control and lighting controls rather than forcing everything into proprietary silos. (itbrief.co.nz)
For most companies, the hard part of artificial intelligence is not the model. It is getting a model to open the right ticket, read the right database, and not unlock the wrong door in a building system. (anthropic.com) A Model Context Protocol is the new layer people are using for that job. Anthropic introduced the Model Context Protocol in November 2024 as an open standard for connecting artificial intelligence assistants to business tools, files, and software environments. (anthropic.com) The easiest way to picture it is a universal port. The official Model Context Protocol documentation says it works like a “USB-C port” for artificial intelligence apps, so one client can plug into many outside systems without a custom adapter for each one. (modelcontextprotocol.io) That changes the old integration math. Instead of building a separate connector for every model and every business system, developers can implement the protocol once and then expose tools, data, and prompts through a standard server. (anthropic.com) (modelcontextprotocol.io) The protocol has a specific shape. The specification describes a client-host-server setup built on JavaScript Object Notation Remote Procedure Call, with session setup, capability negotiation, and explicit server features for resources and tools. (modelcontextprotocol.io 1) (modelcontextprotocol.io 2) That structure is why regulated companies care. A server can expose a narrow tool like “read badge logs” or “change thermostat setpoint” instead of dumping an entire building management system into a chatbot and hoping the model behaves. (modelcontextprotocol.io) (itbrief.co.nz) The shift is no longer just one company’s idea. OpenAI says its Responses application programming interface supports remote Model Context Protocol servers, and Microsoft said at Build 2025 that it was releasing Model Context Protocol servers for Dynamics 365 enterprise resource planning and customer relationship management apps. (openai.com) (microsoft.com) Infrastructure companies are moving in too. Cloudflare launched Model Context Protocol Server Portals in open beta to centralize, secure, and observe connections, which tells you enterprises are already treating these links as a new security perimeter. (blog.cloudflare.com) Security is where the excitement turns into real work. OpenAI’s documentation warns developers to trust any remote Model Context Protocol server they use, and Microsoft says its own enterprise setup is focused on governance, inventory, and secure-by-default controls for agent connections. (developers.openai.com) (microsoft.com) That is why this story is really about plumbing, not magic. If Model Context Protocol becomes the standard way models talk to heating, ventilation, and air conditioning controls, identity systems, finance software, and document stores, companies get more freedom to swap models without rebuilding the whole building around them. (itbrief.co.nz) (openai.com)
