Plain English

Claude Code Glossary

The words you'll hear around Claude Code and AI building - the CLI, MCP servers, agents, skills, hooks, headless mode, IDE integration, context windows, tokens, and more - explained simply, no jargon, with enough detail to actually be useful.

Claude Code: Claude Code by Anthropic is an agentic coding tool that runs in your terminal, inside popular editors like VS Code and JetBrains, and through the Claude desktop app or the web. You describe what you want in plain English and Claude reads, writes, and edits real files in your project, runs shell commands, searches your codebase, and iterates with you - so you can ship working software without hand-writing every line. It is distinct from the regular Claude chat interface: where chat gives you text, Claude Code gives you a working codebase.
Claude Code CLI: The command-line interface that ships with Claude Code. Once you run claude code install (npm install -g @anthropic-ai/claude-code), you get the claude command in your terminal. From there you can run claude to open an interactive session, or pass flags like --print to pipe output into scripts, making it easy to wire Claude Code into CI pipelines, build scripts, and automations. The CLI is the most direct way to use Claude Code and the foundation every other integration builds on.
Claude Code Install: Getting Claude Code onto your machine takes a single npm command: npm install -g @anthropic-ai/claude-code. Node 18 or later is required. After the package installs you run claude code login to authenticate your Anthropic account, and then you can open any project folder and type claude to start a session. The install process is designed to be minimal - there is no GUI to download and no separate runtime to manage.
Claude Code Login: After installing the CLI, claude code login opens your browser and walks you through authenticating with your Anthropic account (or a team API key). The token is stored locally so subsequent sessions start without re-authenticating. Teams can also inject an ANTHROPIC_API_KEY environment variable instead of using the interactive login flow, which is useful for CI servers and containerized environments.
Claude Code Desktop App: Anthropic ships a native desktop application for macOS and Windows that gives you a windowed interface for Claude Code without needing a terminal open. The claude code desktop app supports the same agentic capabilities as the CLI - file editing, command execution, MCP servers - but adds a persistent sidebar, project history, and a more approachable UI for people who are not yet comfortable in a terminal. It is a good starting point when you are first learning Claude Code.
Claude Code on the Web: Claude Code is also available directly in the browser at claude.ai, as part of the Projects feature. Claude Code on the web lets you run agentic sessions without installing anything locally, which makes it easy to try before you commit to a local setup or to do quick tasks from a machine that does not have the CLI installed. File access is handled through file uploads rather than direct filesystem access, so the scope is narrower than the desktop or CLI.
What Is Claude Code: Claude Code is Anthropic's answer to the question of what happens when you give an AI model direct access to your codebase and tools. Instead of you copying code into a chat window, Claude Code sits inside your development environment, reads your actual files, runs your actual commands, and makes precise edits. The result is a workflow where you describe an outcome - 'add user authentication to this Express app' - and Claude Code figures out which files to change, makes the changes, runs your test suite, and fixes any failures. For people learning to build software, it removes the blank-page problem and turns every project into a guided build-along.
Claude: The family of AI models built by Anthropic that powers Claude Code. Different versions - Opus, Sonnet, and Haiku - trade off speed, cost, and reasoning depth. Claude Code defaults to the most capable available model for agentic tasks, but you can specify a model tier if you are cost-sensitive or need faster iteration on small changes. Understanding the model tiers helps you spend your API credits efficiently.
Claude Code by Anthropic: Claude Code is an official, first-party product from Anthropic - the safety-focused AI company founded by Dario Amodei, Daniela Amodei, and former OpenAI researchers. Unlike third-party coding assistants built on top of Claude's API, Claude Code is built and maintained by the same team that builds Claude itself, which means it gets model improvements, new tool-use capabilities, and safety updates directly. When Anthropic ships a new Claude model, Claude Code typically supports it on day one.
Agent: An AI that can take a goal, decide on the steps to reach it, and carry them out using tools - reading files, running commands, calling APIs - with little step-by-step instruction from you. Claude Code is itself a claude code agent: you give it a task and it plans and executes the work autonomously. You can also use Claude Code to build your own custom agents: scripts or services that use the Claude API and tools to automate workflows, process data, or interact with external systems.
Subagent: A separate, focused agent that the main agent can hand a specific task to - for example, exploring a codebase, running a test suite, or making a targeted set of edits - so the work happens in its own isolated context. Subagents let the orchestrating agent parallelize work across multiple concerns without mixing them up in a single long conversation. In Claude Code you can explicitly invoke subagent-style delegation using the Task tool, which launches a fresh context for a specific job and returns the result. Read more →
MCP (Model Context Protocol): An open standard, created by Anthropic and open-sourced to the community, that gives AI tools like Claude Code a consistent way to connect to external data sources and services. Before MCP, every tool integration was a one-off custom implementation. MCP standardizes the protocol so that any compliant server - whether for a database, a browser, a cloud service, or your local file system - can be plugged into any compliant client. Think of it as USB for AI tool integrations: build the server once and it works everywhere the standard is supported.
MCP Server: A small program - often just a few hundred lines - that exposes a specific tool or data source to Claude Code over the Model Context Protocol. Common MCP servers give Claude access to your file system, a running browser (for scraping or testing), GitHub, Linear, Slack, a Postgres database, or external search APIs. You configure MCP servers in a JSON file (claude_desktop_config.json or .mcp.json) and Claude Code discovers them on startup. Claude Code Club members have access to guides covering 25+ MCP servers and when to reach for each one.
Claude Code Skill: A reusable, packaged instruction set you give Claude Code so it knows how to do a specific task your way - your conventions, your stack, your steps, your tools, all bundled together. A claude code skill might define how you write tests, how you structure a new component, or how you do a code review. Instead of explaining your conventions in every new session, you write a skill once and invoke it as needed. Claude Code Club ships 50+ skills members can use as-is or adapt.
Slash command: A shortcut, typed as /name, that expands into a longer instruction or full workflow inside Claude Code. You can define your own slash commands to encode repeated prompts - for example /deploy to run through your deployment checklist, or /review to trigger a code-review skill. Slash commands make individual workflows portable and keep your whole team executing the same playbook. They are stored as Markdown files in your project's .claude/commands directory. Read more →
Hook: A script you configure to run automatically in response to lifecycle events inside Claude Code - for example, before Claude executes a shell command, after it writes a file, or when a session ends. Hooks let you enforce project rules, post-process output, log activity, or trigger downstream actions without having to ask the model to remember to do them. A common use case is a pre-tool hook that blocks Claude from running certain destructive commands in production environments. Read more →
CLAUDE.md: A Markdown file you place in a project (or in your home directory for global rules) where you write standing instructions and context for Claude Code. Your tech stack, coding conventions, folder structure, things-to-never-do, preferred libraries - all of it lives here. Claude reads the file at the start of every session, so you document your project once rather than re-explaining it in every prompt. Teams use CLAUDE.md to onboard Claude the same way they onboard a new developer: write it down and let Claude read it. Read more →
Headless Mode: Running Claude Code without an interactive terminal session - passing a task via flags and receiving structured output. The --print flag (or the equivalent SDK method) is the primary way to use headless mode. It lets you embed Claude Code inside scripts, cron jobs, GitHub Actions, and other automated pipelines where there is no human at the keyboard. Headless mode is how builders turn one-off Claude Code sessions into repeatable automated workflows. Read more →
IDE Integration: Claude Code integrates with VS Code, JetBrains IDEs, and any editor that supports the Language Server Protocol. The integration surfaces Claude Code's capabilities - file editing, running commands, explaining code - directly in the editor's sidebar or command palette, so you never have to leave your editor to describe a change. For people learning to build software, the IDE integration is often the gentlest entry point because it feels like a chat panel attached to the work you are already looking at.
Context window: The maximum amount of text - measured in tokens - that a model can consider in a single pass. This includes your message, the conversation history, every file Claude has read, and every tool result returned so far. Claude models have large context windows, but large codebases can fill them up quickly. Skilled Claude Code users learn to manage context deliberately: use CLAUDE.md for standing context, open only the files that are relevant to the current task, and start a fresh session when the task changes significantly. Read more →
Token: The unit AI models use to measure and price text. Roughly speaking, a token is a word-piece - about 0.75 English words on average. Both your inputs and Claude's outputs count against your token usage, and Claude Code usage is billed per token through your Anthropic account. Understanding tokens helps you estimate costs and manage your context window: a typical source file might be 500-2000 tokens, and a full repository might be hundreds of thousands. Read more →
Prompt: The instruction you give the AI. In Claude Code, a good prompt is specific, concrete, and outcome-focused - it names what done looks like, not just what direction to go. 'Add a login form to this page' is less useful than 'Add a login form to src/pages/Home.tsx that validates email format client-side and calls POST /api/auth/login on submit.' The skill of writing good prompts is the single biggest lever on the quality of what you get back, and it is a core part of what Claude Code Club teaches.
System prompt: The behind-the-scenes instructions that configure an AI's role, persona, constraints, and behavior before the conversation begins. In Claude Code's architecture the CLAUDE.md file acts as a form of persistent system context: it is injected into every session so Claude always has your project's background. When you build your own agents on top of the Claude API, you write a system prompt explicitly to define how the agent behaves, what tools it can use, and what it should refuse to do. Read more →
Tool use: When an AI model calls an external function - running a shell command, reading or writing a file, querying a database, searching the web - instead of only producing text. Tool use is what separates a chat interface from a real claude code agent. Claude Code ships with a set of built-in tools (read_file, write_file, bash, etc.) and can acquire additional tools through MCP servers. Every time Claude Code edits a file or runs a test, it is using tool use under the hood. Read more →
Vibe coding: Building software by describing what you want to an AI in natural language and iterating on the results, rather than writing every line by hand. The term emerged in 2024-2025 as agentic coding tools like Claude Code made it practical to go from idea to working app in hours instead of days. Vibe coding is a fast path to working software - and the people who do it most effectively also know how to read and review what the AI produces so they can catch mistakes before they compound. Read more →
Prompt engineering: The practice of writing and structuring prompts to reliably get the output you want from an AI model. For agentic tools like Claude Code, prompt engineering is less about magic wording and more about giving the model clear goals, relevant context, concrete constraints, and examples of what good output looks like. Breaking a large goal into smaller, well-scoped tasks - rather than asking Claude to 'build my whole app' - is the single most reliable prompt engineering technique for getting quality results. Read more →
Anthropic: The AI safety company that built Claude and Claude Code. Founded in 2021 by former OpenAI researchers, Anthropic focuses on building AI systems that are safe, steerable, and interpretable. Claude Code is one of Anthropic's flagship products, designed to make the power of Claude models accessible directly inside a developer's workflow. Anthropic publishes extensive documentation on how to use Claude Code, and the team actively maintains and updates the tool with each new model release.
API key: A secret string that authenticates your requests to Anthropic's servers. When you use Claude Code, either through claude code login (which handles authentication interactively) or by setting ANTHROPIC_API_KEY in your environment, every request the tool makes is signed with this key. Your API key is tied to your Anthropic account and its associated billing. Keep it secret - never commit it to a public repo - and rotate it if it is ever exposed.
Agentic loop: The core execution cycle of an AI agent: receive a goal, plan a step, use a tool, observe the result, plan the next step, repeat until the goal is reached or the agent decides to ask for help. Claude Code runs an agentic loop on every non-trivial task - it does not just write code and stop; it runs the tests, reads the error output, fixes the failures, and checks the result again. Understanding the agentic loop helps you write tasks that play to this strength: describe an outcome you want checked, not just a change you want made. Read more →
Workspace: The project directory Claude Code is operating in. Claude Code treats the workspace as its source of truth: it reads the directory tree, opens files, and makes edits all within this folder (and sub-folders). You can configure per-workspace settings in a .claude directory at the project root, including a local CLAUDE.md, custom slash commands, and MCP server overrides. Keeping your workspace well-organized - clear folder structure, a good CLAUDE.md - has a direct impact on how reliably Claude Code navigates your project.

Stop Googling the jargon. Start building.

We'll teach you every one of these hands-on. Join Claude Code Club for $9/month.

Join the club See the curriculum

Related: curriculum, guides, and the blog.