Claude: Zero to Hero
Power user · module 7 of 8

Plugins

You are here · Paths A, B, C · Requires Pro or above · 25–35 min · Assumes Skills and Connectors.

What you’ll learn: what a plugin bundles, where to find good ones, and when to build your own.

If you only read one thing

A plugin is a bundle — several Skills, connectors and helpers packaged together for a particular role or tool, installed in one click instead of assembled by hand. If Skills are single recipes, a plugin is the cookbook.

You’ll mostly consume them rather than build them: browse the directory, install one matching your work, see what it gives you. Building your own only makes sense once you have several Skills you want to distribute to other people as a set.

Same security rule as Skills, more so, because there’s more inside: a plugin can give Claude instructions and access. Install ones from sources you’d trust with software.

What a plugin is

A plugin is a single installable package that can contain:

The point is distribution. Everything above can be set up individually; a plugin makes “here’s our team’s whole Claude setup” one install.

Plugins work in Claude Code and in Cowork.

Installing one

In Cowork: find plugins in the plugin directory. Install, and the bundled skills, connectors and subagents become available.

In Claude Code: plugins come from marketplaces. Add a marketplace, browse, install. /plugin list shows what’s installed. Plugins can also be loaded from .zip archives and URLs.

Note: plugins that include local MCP servers work through the desktop app only in Cowork.

Namespacing

Plugin skills are namespaced — /my-plugin:review rather than /review — so multiple plugins can coexist without clobbering each other. This matters more than it sounds once you have several installed.

What good plugins exist

Anthropic and the community publish plugins for common needs. Two worth knowing:

security-guidance — has Claude review its own code changes for vulnerabilities and fix them in the same session. If you write code with Claude and ship it, install this. See Catch security issues as Claude writes code.

Code intelligence plugins — connect Claude to a language server for your language, giving symbol-level navigation and live type errors instead of grep. Substantially better on large typed codebases. See Discover plugins.

Browse everything in the skills, connectors and plugins directory.

When to build your own

Build a plugin when a second person or a second repository needs the same setup.

Before that point, individual skills and MCP configs are simpler. The progression is:

Correct Claude in chat
  → the same correction twice → write it into instructions / CLAUDE.md
  → the same procedure three times → make it a Skill
  → a second repo needs it → make it a Plugin
  → a second team needs it → publish to a Marketplace

Don’t skip steps. A plugin built before you’ve used the underlying skills for real is a plugin full of guesses.

Building plugins is covered in Plugins and marketplaces.

Security

Same rules as Skills, but more so — a plugin can bundle MCP servers and hooks, which is a broader capability surface than instructions alone.

Try it

Exercise 1 — Browse. Open the plugin directory. Read the descriptions of ten plugins. Note which two are relevant to your actual work.

Exercise 2 — Install one, deliberately. Install a single plugin. Before you do, write down what you expect it to add. After, check /plugin list (Claude Code) or the plugin’s page and see whether you were right.

Exercise 3 — Audit. For a plugin you’re considering, enumerate: skills, subagents, hooks, MCP servers. For each hook, what command does it run? If you can’t find out, don’t install it.

Exercise 4 — Plan your own. Write down what would go in a plugin for your team. Skills? MCP servers? A hook? You don’t have to build it — but knowing the shape tells you what to build first.

Checkpoint

Going deeper