Claude Code is more than a terminal chat. It's an extensible agent platform, and people keep searching for the same set of primitives: skills, slash commands, hooks, subagents, CLAUDE.md, MCP servers, channels, and plugins. This hub explains what each one does, when to reach for it, and links to the full build guide for every layer.
The short answer
Think of Claude Code as a 4-layer extension stack: CLAUDE.md (project memory), skills (reusable, auto-invoked capabilities), hooks (rules that fire on events), and subagents (separate agents with their own context windows). On top of that, MCP servers connect Claude Code to external tools, channels push notifications into your terminal, and plugins package and share the whole setup. Start with CLAUDE.md and one skill; add the rest as you need them.
Skills, slash commands, hooks and CLAUDE.md
These are Claude Code's core extension layers. CLAUDE.md holds your project memory and rules. Skills are reusable capabilities Claude auto-invokes when relevant. Hooks fire rules on events (before/after a tool runs). Slash commands are saved prompts you trigger by name. Together they form a 4-layer stack you can build in about 20 minutes, with a decision framework for picking the right layer.
Read the full breakdown: Claude Code Skills, Commands, Hooks & Agents Guide (and What Goes in CLAUDE.md)
Subagents
Subagents are separate Claude Code agents, each with its own context window, that run in parallel. A common pattern is a 3-agent research team (researcher, writer, reviewer) working at the same time while you do something else. Use subagents when one context window isn't enough or when tasks can run independently, rather than cramming everything into a single agent.
Read the full breakdown: I Built a 3-Agent Research Team in Claude Code (They Run in Parallel)
MCP servers
MCP (Model Context Protocol) servers connect Claude Code to outside tools and data, like Perplexity for web search, Firecrawl for scraping, or Notion for your docs. Pair them with automation hooks (formatting, safety, quality checks) so Claude Code can both reach external systems and enforce your rules. MCP is how you extend Claude Code beyond your local files.
Read the full breakdown: How to Set Up MCP Servers and Hooks in Claude Code
Channels (Telegram and Discord)
Claude Code Channels push webhooks, alerts, and chat messages between Claude Code and apps like Telegram and Discord, so you can drive your terminal from your phone or get notified when a long task finishes. The full guide walks through Telegram, Discord, and custom webhook setup in about 30 minutes, including the channel code.
Read the full breakdown: Claude Code Channels Setup Guide: Telegram, Discord, and Custom Webhooks
Plugins
Plugins package your Claude Code setup (skills, commands, hooks) so you can reuse and share it. The structure most tutorials get wrong: a correct .claude-plugin/plugin.json, the Claude Agent SDK, GitHub Actions v1, and a marketplace install path. Build a plugin once your config is worth reusing across projects or distributing to others.
Read the full breakdown: How to Create a Claude Code Plugin (and Why Most Tutorials Get the Structure Wrong)
Go deeper
- Train Claude Code to write in your brand voice
- Build a technical content verification subagent (Build Log #5)
- Build a research AI agents automation in Claude Code
- Turn Claude Code into a full content system
- Browse all Claude Code tutorials
Frequently asked questions
What is CLAUDE.md in Claude Code?
CLAUDE.md is Claude Code's project memory file. It holds your instructions, conventions, and context so Claude follows your rules automatically on every task. It's the base layer of the 4-layer extension stack, beneath skills, hooks, and subagents.
What's the difference between a Claude Code skill and a subagent?
A skill is a reusable capability Claude auto-invokes inside the current session when it's relevant. A subagent is a separate agent with its own context window that can run in parallel. Use a skill for reusable logic, a subagent when you need isolation or concurrency.
How do I connect Claude Code to Telegram or Discord?
Use Claude Code Channels, which push webhooks, alerts, and chat messages between Claude Code and apps like Telegram and Discord. The setup (Telegram, Discord, or custom webhooks) takes about 30 minutes and lets you drive your terminal from your phone or get notified when tasks finish.
What are MCP servers in Claude Code?
MCP (Model Context Protocol) servers connect Claude Code to external tools and data, such as Perplexity for search, Firecrawl for scraping, or Notion for documents. They are how you extend Claude Code beyond your local files.
Browse all Claude Code tutorials · Build AI agents with the Agents Toolkit