What Problems Do Plugins, Connectors, and MCP Solve?
Choose an extension by the missing capability: packaged behavior, access to external data, or a protocol that exposes tools and resources.
A Skill defines how AI should perform a task. A connector or app links AI to an existing service. MCP provides a standard interface for exposing data and tools. A plugin usually packages installable capabilities. Product names change, so identify the missing capability before choosing a label.
Suppose you want AI to organize Drive files, query a CRM, and create a summary in a fixed format. That request mixes a work method, an external data connection, and actions the system may execute. Installing more extensions does not clarify those boundaries.
OpenAI documents how ChatGPT apps connect data and actions, while the MCP specification defines a standardized interface. The four-way map in this lesson is my cross-product teaching method. Product names and contents must still be checked against current official documentation.
A plugin packages a set of capabilities
“Plugin” generally refers to an installable or enabled package of capabilities. It may include instructions, Skills, app connections, tools, or an interface. The word does not refer to one universal technical standard across products.
Ask, “Do I need a packaged capability that is distributed and updated together?” Do not ask whether a plugin is inherently more powerful than a Skill.
A connector or app links an existing service
When data already lives in Drive, Slack, a CRM, or another SaaS product, a connector or app lets an AI system access that service. Depending on the connection, it may search and read data or create, modify, send, and delete content.
OpenAI states that an app may receive conversation context relevant to a request, and that third-party terms and permissions also apply. OpenAI: Apps in ChatGPT
Before connecting, ask whose account is used, which data can be read, which write actions are available, where the data flows, and which actions require approval.
MCP standardizes how AI applications receive context and tools
Model Context Protocol (MCP) is an open protocol for connecting LLM applications with external data sources and tools. The current specification lists resources, prompts, and tools among the main capabilities a server may provide: contextual data, templated interactions, and executable functions. MCP Specification
MCP standardizes an interface. It does not prove that a server is safe, that its data is correct, or that its permissions suit your task. Before connecting an unfamiliar MCP server, check the provider, trust basis, exposed tools, and data flow.
Choose by the missing capability
| Missing capability | Start by examining |
|---|---|
| A reusable work method | Skill |
| A connection to an existing service | Connector or app |
| A standard interface for data and tools across applications | MCP |
| An installable package of capabilities | Plugin |
The categories can overlap. A plugin may contain a Skill and depend on an app. A Skill may call an MCP tool. The actual boundary comes from the product and package you are using.
Practice: complete an extension choice record
Work goal:
Currently missing: method / data / external action / packaged capability
Service or data to connect:
Read only, or write access too?
Account and permission used:
Provider and trust basis:
Where data will go:
Approval and stop conditions:
Candidate: Skill / connector-app / MCP / plugin / do not connect yetIf you cannot explain the data and actions, do not connect the extension yet.
The next lesson introduces Agent work. When AI can choose tools, execute several steps, and adjust based on results, permissions and supervision become more important than the extension’s name.