"This is apparently the plumbing. Nobody mentioned the plumbing."— Office Man
What is MCP?
Model Context Protocol. An open standard developed by Anthropic, released in late 2024, that describes a consistent way for AI tools to connect to external systems — files, databases, APIs, applications. Before MCP, every connection between an AI tool and an external system was a custom engineering project. With MCP, a standard connection means that any MCP-compatible AI tool can connect to any MCP-compatible system without rebuilding the integration from scratch. The USB analogy is useful: before USB, every device needed its own connector; after USB, compatibility was built into the standard. MCP is attempting something similar for AI integrations.
Why does this matter?
The most useful things AI tools can do in a professional context involve connecting to real systems — reading documents from SharePoint, querying a database, checking a CRM record, retrieving information from a calendar or project management tool. Without a common standard for building those connections, each one is expensive, slow to build, and difficult to maintain. MCP makes integrations faster and cheaper to create, easier to keep up to date, and more consistent in how they are secured. It also opens the ecosystem: a smaller AI tool can now connect to the same systems as a large one, if both support the standard.
Who uses it and how?
Developers and IT teams building or deploying AI tools. The end-user experience is not "I am using MCP" — it is "this AI tool can search our internal documents" or "this assistant can look up a customer record". MCP is the mechanism underneath that makes the connection work. It is also the plumbing that will enable more capable AI agents — tools that carry out multi-step tasks involving several different systems in sequence. Those kinds of workflows require reliable, maintainable connections to real data. MCP provides the standard for building them.
What to do
If you are evaluating AI tools for integration work, ask whether they support MCP — it is an increasingly relevant question. If you are in IT or building AI tools, read Anthropic's public documentation at modelcontextprotocol.io. If you are an end user, you do not need to do anything specific — but knowing the name helps you ask better questions when someone describes an AI tool connecting to your systems. And if you are procuring tools that need to connect to internal data, include integration approach and MCP support in your evaluation criteria.