Model Context Protocol (MCP)
An open standard for connecting AI tools to other software and data, so a connector built once works in every AI product that supports the standard.
Also called: MCP · MCP server · MCP connector · custom connector
Connections
Understand first
Ideas this one builds on.
Evolved from
How the idea changed over time.
Resources that teach it
Competency
Find model context protocol (mcp) on the commons map ↗
You may know it from
- hardware USB-C One shared plug shape means any device can connect to any charger or screen that supports it.
- logistics the standard shipping container A shared box size let ships, trains and trucks handle any cargo without repacking.
In products
The same idea, under different names in different tools.
| Provider | What they call it |
|---|---|
| Anthropic | MCP connectors in Claude (Anthropic started the standard) |
| OpenAI | MCP support in ChatGPT developer mode and the Agents SDK |
| MCP support in Gemini CLI |
In plain words
Before MCP, every AI product needed its own custom-built connection to every other app. A connector built for one chatbot didn’t work in another. Early ChatGPT plugins were one such attempt, later retired.
MCP, introduced by Anthropic in late 2024 and since adopted by other major AI companies, is a shared format for those connections. A company or a developer writes one MCP server for their app, and any AI tool that speaks MCP can use it.
You rarely need to touch MCP directly. It matters for nontechnical users in two ways. It explains why your favourite app may suddenly appear as a connector in several AI tools. And it means you should judge each connector by who built it and what access it asks for, as you would a browser extension.
Example
An events team uses a project-management app that offers an MCP server. Their admin adds it as a custom connector, and staff can now ask their AI tool “what’s overdue on the gala board?” without exporting anything.
How it connects
- A standard way to build tools and connectors, which it evolved from.
- Connectors built this way give agents their reach into real systems.
- Each server you add is a privacy and data choice.
Try it
Look at the connector list in your AI tool. Guess which ones were built by the app’s own company and which by third parties, then check two of them. Note what access each one asks for.
Read this card as plain Markdown: /concepts/con-mcp.md. Know a parallel from your own field, or a product that implements this? Add it.