MCP, Skills, Agents, Tokens, Context — explained in plain English with real examples.
MCP is a standard way for AI apps to connect to external tools and data.
ExampleWith a GitHub MCP server connected, an AI app may be able to use repository tools to read issues, pull requests or repository information.
A Skill is a reusable set of instructions and resources that helps an AI agent handle a certain kind of task.
ExampleA YouTube Skill might define a repeatable flow for analyzing a topic, proposing titles, writing a description, preparing tags and checking the final format.
An AI agent can plan steps, use tools and take actions toward a goal instead of only producing one reply.
ExampleAsked to add an English version of a website, a coding agent might inspect the project, change routes, build the site, test links and fix errors.
A token is a basic unit an AI model uses to process content; it may be a character, part of a word, a full word or punctuation.
ExampleSaying an MCP tool uses many tokens may mean its tool descriptions, context or returned content occupy a large amount of the model's working space.
Context is the information an AI can refer to while handling the current task.
ExampleWhen Codex edits a website, your request, project files, AGENTS.md, loaded Skills and tool definitions may all become context.
A CLI lets you operate software by entering text commands in a terminal instead of clicking buttons in a graphical interface.
ExampleRun gh issue list to inspect GitHub issues without opening the site and clicking through its interface.
An API is an agreed interface through which software can exchange data or call another system's capabilities.
ExampleA weather app can request today's forecast from a weather service API and display the returned data.
RAG lets an AI retrieve relevant information first, then answer using that information.
ExampleWhen an employee asks about travel reimbursement, a RAG system can retrieve the relevant company policy before the model answers.