Skip to content
MCP Ready

The terminology guide

MCP is the standard. “App” and “connector” are usually the packaging.

Vendors choose product language that makes sense in their interface. That is why one MCP-powered integration may appear as an app, another as a connector, and another as a plugin.

The short translation

Comparison details
TermPlain-English meaningImportant distinction
MCPThe open communication standard.It is a protocol, not a product or hosted service.
MCP serverThe software that makes a focused set of capabilities available through MCP.It can be local or remote, internal or public.
ToolOne structured operation the AI can request.A tool is part of a server, not the whole integration.
ConnectorA configured or vendor-maintained bridge to another system.Product term; it may use MCP underneath.
App / integrationThe user-facing package or connection people enable.May include tools, authentication, branding, and optional UI.
PluginA distributable package of capabilities in a product ecosystem.Depending on the vendor, it may bundle an MCP server, skills, UI, or templates.
MCP AppAn optional standard extension for interactive HTML UI returned by an MCP server.It is not a generic synonym for every MCP integration.
Host / clientThe AI application and its protocol connection component.The host is what the user experiences; the client is mostly internal plumbing.

One stack, several labels

User sees

App, connector, plugin, or integration

→

Package contains

MCP server, tools, auth, and optional UI

→

Connects to

APIs, databases, files, and business systems

Why the same thing gets renamed

Protocol language is optimized for implementers. Product language is optimized for people choosing and enabling features. A store may list an “app,” an administrator may approve a “connector,” and a developer may paste an “MCP server URL”—all while discussing different layers of one implementation.

Names also change as products mature. Treat the label in a menu as a clue, then ask four practical questions:

  1. 01

    What can it read?

    Which accounts, files, records, or external sources can become available?

  2. 02

    What can it change?

    Can it create, send, publish, update, delete, or spend?

  3. 03

    Who authorizes it?

    A person, a workspace administrator, a shared service account, or several of these?

  4. 04

    Where does it run?

    On one device, inside the company network, or as a third-party remote service?

A few vendor-specific nuances

OpenAI now uses plugin for an installable package, app for an external integration, and MCP server for the endpoint that exposes tools. “Connector” remains in older ChatGPT language and in the API for maintained service connections. See the ChatGPT guide.

Anthropic uses “connector” for a connection to a remote MCP server, while local MCP servers, desktop extensions, and plugins have their own packaging and deployment models. See the Claude guide.

MCP Apps is the open extension for interactive interfaces in compatible hosts. It is narrower than the general phrase “an app that happens to use MCP.”

The same idea appears under different names in other products. Our AI product guides compare Microsoft Copilot, Cursor, GitHub Copilot, and Google Gemini as well.

A durable way to say it

“This app connects our AI workspace to the CRM through an MCP server. The server exposes separate tools for finding accounts and preparing approved updates.”

Terminology verified against the MCP specification, OpenAI plugin and MCP documentation, and Anthropic custom connector documentation in July 2026.