Discovering and Installing Plugins

Level: 🌱 Beginner Related: Plugins, Plugin Marketplaces

The Scenario

You’ve been using Claude Code for a few weeks. You’re comfortable with the basics — asking questions, editing files, running commands. But you’ve heard about plugins and haven’t explored them yet. Today, you decide to see what’s available and set up your first plugins.

The Walkthrough

Step 1: Opening the Plugin Browser

You type /plugin and see the plugin browser for the first time. There are two tabs: Installed (empty) and Discover. You switch to Discover and see a list of available plugins from Anthropic’s official marketplace.

Each plugin shows its name, a short description, who made it, and what components it includes (skills, agents, hooks, MCP servers). You scroll through and see plugins for code review, frontend design, language servers, and more.

What’s Happening: The Marketplace ecosystem has multiple tiers. The official Anthropic marketplace is available by default — no setup needed. Community marketplaces can be added later for more options.

Step 2: Your First Plugin — frontend-design

You’ve been building a React dashboard this week and Claude’s UI suggestions have been decent but generic. You notice frontend-design in the list: “Improves Claude’s frontend and UI work.” It’s from Anthropic, and it contains a single auto-invoked skill. Low commitment.

You select it and choose User scope — you want it available across all your projects, not just this one. Claude installs it and you restart your session.

Now when you ask Claude to build a component, the responses are more design-aware — better spacing, accessibility considerations, responsive patterns. You didn’t invoke anything — the skill’s description matched your frontend prompts and activated automatically.

What’s Happening: Skills with a description field can be auto-invoked when the description matches what you’re asking about. This is the simplest plugin pattern — a single skill that enhances Claude’s behavior in a domain without requiring you to learn new commands.

Step 3: A Heavier Plugin — code-review

Your team has been doing manual code reviews and you want Claude to help. You find code-review in the Discover tab. This one is more substantial: it includes 5 parallel agents (correctness, security, performance, maintainability, test coverage) plus orchestrating skills.

You install this one at Project scope — it goes into .claude/settings.json so your teammates get it too when they pull. After restarting, you test it on a recent PR branch. Claude spawns five agents that work in parallel, each focused on a different review dimension. The results come back as a unified review document.

What’s Happening: Plugins can be installed at different scopes. User scope (in ~/.claude/settings.json) is personal and global. Project scope (in .claude/settings.json) is shared with the team via version control. Choose the scope that matches who should have the plugin.

Step 4: Building Your Own — plugin-dev

You’ve now installed two plugins and seen how they work. The claude-code-guide plugin you installed earlier was just markdown files — no application code at all. You think: “I could build something like this for my team’s deployment runbook.”

Back in the Discover tab, you find plugin-dev from Anthropic. It’s a guided plugin creation workflow — skills, agents, hooks, structure, validation, the whole process. You install it.

/plugin install plugin-dev@claude-plugins-official

Now you can run /plugin-dev:create-plugin and Claude walks you through scaffolding a new plugin from scratch: choosing components, writing the manifest, creating skills, and testing with --plugin-dir. Within an hour you have a working deployment-runbook plugin ready to share with your team.

What’s Happening: Plugins can help you build more plugins. The plugin-dev plugin demonstrates the meta-plugin pattern — using Claude’s own extension system to guide creation of new extensions. See Plugins for the full manual approach to plugin creation.

Step 5: An MCP Plugin — context7

Your current project uses several libraries with fast release cycles, and Claude sometimes suggests outdated APIs. Still in the Discover tab, you find context7 from Upstash — it provides current library documentation via MCP. You notice it’s marked as community-managed, meaning Upstash maintains it rather than Anthropic.

You check the description and components: MCP server, skills, agents, and commands. It’s more substantial than a skills-only plugin, so you read the README before installing. Everything looks solid — active maintenance and clear documentation.

You install it and try asking about a library. Now Claude fetches the latest docs instead of relying on training data. The difference is immediate — function signatures match the version you’re actually using.

/plugin install context7@claude-plugins-official

What’s Happening: MCP servers connect Claude to external data sources. The context7 plugin bundles an MCP server that fetches library documentation on demand. Community-managed plugins on the official marketplace are maintained by their authors — check their documentation and activity before installing. See Plugin Examples for more examples of the MCP-in-a-plugin pattern.

Step 6: Taking Stock

You check your plugin setup:

/plugin

On the Installed tab, you see:

Plugin	Scope	Components	Why you installed it
frontend-design	User	Skill (auto-invoked)	Better UI/design output across all projects
code-review	Project	Agents, Skills	Team code review workflow for this project
plugin-dev	User	Skills	Guided plugin creation workflow
context7	User	MCP, Skills, Agents	Up-to-date library docs across all projects

The pattern: lightweight personal tools at User scope, team workflow tools at Project scope. If you ever need to disable something, it’s a quick /plugin disable away without uninstalling.

What’s Happening: Plugin management is designed to be low-friction. Install, disable, re-enable, uninstall — all from /plugin. See Plugins for the full management commands.

Key Takeaways

Start with /plugin → Discover tab. Browse what’s available before committing to anything.
Install lightweight plugins at User scope. Skills-only plugins like frontend-design are low-risk and broadly useful.
Install team workflow plugins at Project scope. Heavier plugins like code-review benefit from shared configuration.
Add community marketplaces for more options. But evaluate community plugins before installing — check the author, components, and documentation.
MCP plugins connect Claude to live data. If Claude’s training data is insufficient for your libraries, MCP plugins like context7 bridge the gap.
You can always disable or uninstall. Plugin management is reversible, so experiment freely.