K-Culture GlossaryMWords from the people who build
MCP
Model Context Protocol
A common standard for plugging AI into outside tools and data — nicknamed the "USB-C of AI."
In plain words
MCP (Model Context Protocol) is a common standard for connecting AI to external tools and data. Anthropic released it in late 2024, and it became an industry standard after OpenAI and Google adopted it too. Its nickname, "the USB-C of AI," is less a metaphor than a near-literal description.
Before MCP, hooking up an AI to even one tool meant building a custom connection from scratch every time — like the era when every device had its own charging port. MCP unified the port. Build a single "MCP server" that follows the spec, and it plugs into any AI that speaks the protocol, whether that's Claude, ChatGPT, or others.
This matters because it's the infrastructure underneath the AI agent era. For an AI to read your calendar, search company documents, or even make payments, it needs to connect to tools — and MCP is the wiring standard that makes that possible. When you see a headline like "Service X now supports MCP," read it as "AI can now be plugged into that service."
How it shows up in the news
"With all three major AI companies now officially supporting MCP, it has effectively become the standard" — much like the unification of charging ports, this kind of milestone tends to be followed by ongoing news of which services have connected via MCP.
Try it yourself
- Look for "Connectors" in the Claude desktop app or ChatGPT settings — these connectors work through MCP.
- Try connecting one, like Google Drive or Google Calendar.
- Ask something like, "Find last month's meeting notes in my Drive and summarize them." That reach into outside data — that's MCP doing the wiring.
API vs. MCP: What's the Difference
A shared standard AI uses on its own, A window developers wire up by hand
Both get described as "connecting AI to the outside world," so they sound like two names for the same thing. But an API is a one-off window that differs from service to service, while MCP is an agreement to make all those windows the same shape. They sit at different layers — even a service that supports MCP is using an API underneath.
| Aspect | MCP | API |
|---|---|---|
| What it is | A shared standard that unifies the shape of those windows | A separate ordering window for each service |
| Who uses it | The AI itself picks and uses connected tools | A developer writes code to call it |
| Direction of the call | Usually AI calling on outside tools | Usually an app calling on AI |
| Cost of hooking up | Build to the standard once, and it plugs into any AI | 100 services means building 100 separate connections |
| Analogy | The USB-C connector standard | A restaurant's order counter |
| In articles | Comes up with connection talk like agents and connectors | Comes up with money talk like price cuts and usage |
Rule of thumbIf it's a story about an app calling on AI, it's API. If it's a story about AI calling on your tools and data, it's MCP.
MCP vs Skills: what's the difference
Does it plug in a tool, Does it teach a way of working
Both keep showing up together in agent articles, and both get described as 'adding something to AI,' so they look alike. But one attaches a hand, and the other hands over a manual. A manual without a hand can't act on what it reads; a hand without a manual won't do things our way.
| Aspect | MCP | Skill |
|---|---|---|
| What it provides | A connection that reaches outside tools and data | A procedure for doing that work our way |
| What it is | An industry-shared standard | A folder holding instructions and reference material |
| Who makes it | The company that owns the service opens up a server | The user writes it themselves — just needs to be able to write |
| Examples | Connecting to Google Drive, Calendar, internal documents | Our company's quote template, our team's announcement tone |
| Analogy | A tool and a key | A work manual |
Rule of thumbIf the AI can't act because there's nothing to reach for, MCP is what's needed; if it already knows how but not in our way, a Skill is what's needed.
See also
Stories using this term
No story has used this term yet. New ones attach here automatically.