
Model Context Protocol (MCP): a USB-C port for AI tools
In short: MCP is an open protocol that lets AI apps connect to tools and data through standard servers: what it is, how it works and when you need it.
Model Context Protocol (MCP) is an open protocol, introduced by Anthropic in November 2024, that standardises how AI applications connect to external tools and data. A tool provider builds one MCP server; any MCP-capable client, such as Claude, ChatGPT, Cursor or VS Code, can then use its tools, resources and prompts without custom integration code for each app.
Model Context Protocol in plain words
Before USB, every device had its own cable. MCP is the common port: build the plug once, and it fits every AI app that supports it. The project's own documentation makes the same USB-C comparison.
How it works
MCP has three roles. The host is the AI app. The client inside it keeps a connection to each server. Servers expose:
- Tools: actions the model can call.
- Resources: data the app can read into context.
- Prompts: reusable templates.
Servers run locally over standard input/output (stdio) or remotely over Streamable HTTP. Authorization is optional in the specification; when a remote server needs it, the flow is based on OAuth 2.1.
Example
Adding GitHub's remote MCP server to Claude Code from the terminal, as shown in the Claude Code MCP documentation:
claude mcp add --transport http github https://api.githubcopilot.com/mcp/ \
--header "Authorization: Bearer YOUR_GITHUB_PAT"
Replace the placeholder with a GitHub personal access token. After that, the agent can list issues or open pull requests through the server's tools.
Where you'll meet it
- Connectors in Claude, ChatGPT and other assistants
- Coding agents and IDEs such as Claude Code, Cursor and VS Code
- SaaS vendors publishing official MCP servers
- Internal platforms exposing company systems to agents
What it changes for you
For beginners: you need MCP only when you want an assistant to reach a system it cannot already reach. For developers, MCP turns "integrate with every AI app" into "ship one server". For security teams, each server is a new trust boundary that can read data and take actions.
MCP vs an API
An API is how your system exposes functions to code. An MCP server wraps those functions with descriptions and a standard format so a model can discover and call them.
Common mistake
My view: installing community MCP servers like browser extensions. A server runs with your credentials and its tool descriptions go straight into the model's context. Review it as you would a dependency.
Related terms
MCP server · Tool calling · WebMCP · A2A protocol · Prompt injection
From the archive: Stateless MCP moves the state problem onto your side
One MCP server reaches every AI app, so review it like any dependency that holds your credentials.


