MCP, explained: a common plug for AI tools
People ask us whether they should “support MCP”, usually before anyone has said what it is. Here is the short version.
The problem it solves
An AI app is only useful if it can reach your calendar, your tickets, your database, your files. Without a shared standard, every app needs its own connector to every tool. With a handful of apps and a handful of tools, that is already a lot of glue. With a few dozen of each, it is a maintenance job nobody wants.
Figure 1 · Why a standard helps
| Apps | Tools | Custom connectors | With a shared protocol |
|---|---|---|---|
| 3 | 3 | 9 | 6 |
| 10 | 20 | 200 | 30 |
| 25 | 60 | 1,500 | 85 |
What MCP is
The Model Context Protocol is an open standard for connecting AI applications to outside systems. Anthropic introduced it, and described it in December 2025 as a universal standard that had gained wide adoption in its first year. A tool provider writes one MCP server. Any app that speaks MCP can then use it.
Where it lives now
On 9 December 2025, Anthropic donated MCP to the Agentic AI Foundation, a directed fund under the Linux Foundation. Anthropic, Block and OpenAI co-founded the foundation, with support from Google, Microsoft, AWS, Cloudflare and Bloomberg. Block’s goose and OpenAI’s AGENTS.md were contributed at the same time. The point of the move was to keep the protocol open and vendor-neutral.
What it does not do
MCP standardises the connection. It does not decide whether the connection is safe. Every tool you plug in is something the agent could be talked into misusing, so permissions still matter: give each connection the least access it needs, and put a person in front of anything you cannot undo. We wrote about where that review belongs in our note on human review.
Do you need it?
If you are building agents that will touch several systems, or you expect to swap models and apps over time, a shared protocol saves real work. If you are wiring up one integration, a plain API call is fine.