Claude Code plugins: what they are, how to use them safely

What a Claude Code plugin is

A plugin saves setup time: one command installs a setup someone else built, and its updates come from one place.

Under the hood, it's a folder that Claude Code installs and loads as one unit. Inside can be:

• Skills: instructions in a SKILL.md file. Claude loads one when it fits the task, or you run it by name.
• Commands: the older form of a skill.
• Agents: helpers with their own instructions that Claude can hand a task to.
• Hooks: commands Claude Code runs automatically at set moments, such as after every edit.
• MCP servers: connectors that give Claude tools for another system, such as a database.

Plugins come from a marketplace: a catalog that lists plugins and where to download each one.

Know this before your first install: an enabled plugin is part of every session, even ones where you don't use it. Its skill descriptions go along with every message, counting toward your usage, and whatever it runs, it runs as you.

What plugins are good for in production

In production you want the same checks on every change and mistakes caught before customers see them. These kinds of plugins help (categories, not picks):

• Checks after every edit. Hooks that format or lint the code (check it for common mistakes) each time Claude changes a file, whether or not Claude remembers to.
• A second reviewer. Agents and skills that review a change for style, missing tests or security problems before you merge it.
• Pull-request helpers. Commands that commit, push and open a pull request: a request to merge a change, reviewed first.
• Code intelligence. Language servers that show Claude errors right after each edit and let it jump to a definition instead of reading whole files.
• Connectors. Plugins from tool vendors that connect Claude to their service, so it reads real data instead of you pasting it in. Give each the least access that works.

The plugin that can reach your production data is the one to check most carefully.

Plugin or skill: use the smallest thing that works

Every part you add is something to read, update and often pay for in usage, so use the smallest thing that does the job. Skills, agents, hooks and MCP servers all work without a plugin; a plugin is the packaging.

A plain skill is enough when Claude needs instructions: a release checklist, your review rules, a debugging playbook. Commit it in the project's .claude/skills/ folder and your team gets it too.

A plugin is worth it when several parts belong together, say a skill, a hook and a connector, and many projects or a whole team need them.

I prefer plain-Markdown skills, just text files, wherever they do the job, and I read every file of anything I install. My free MVP System Design skill is built that way: 16 chapters of plain instructions for the AI, no code to run. Rules Claude needs in every session belong in CLAUDE.md, not in a skill.

How to install a Claude Code plugin

Anthropic's official marketplace is already there: Claude Code adds it the first time you start it in the terminal. Commands starting with / go into Claude Code; commands starting with claude go into your terminal.

1. Start Claude Code in your project and type /plugin. The Discover tab lists plugins from your marketplaces; type to search.
2. Or open one by name: /plugin install <plugin>@<marketplace>, as in Anthropic's example, /plugin install commit-commands@claude-plugins-official. Nothing installs yet: you see its details first.
3. Read Will install: the skills, commands, agents, hooks and servers the plugin adds.
4. Choose a scope, which decides who gets the plugin:
   • User: you, in every project.
   • Project: everyone in this repository (the project's code and history). Each teammate still installs it once.
   • Local: only you, only here. The smallest reach: start with this.
5. Check that it worked: type / and look for /plugin-name:skill-name, or run /plugin list.

In a local session in the desktop app's Code tab, click + next to the prompt box and choose Add plugins (or Plugins, then Browse plugins).

All commands here are from Anthropic's documentation on plugins, checked on 7 November 2026.

How to add a plugin marketplace

Many plugins sit in their authors' own marketplaces, usually a GitHub repository. Add the marketplace once, then install from it:

/plugin marketplace add <owner>/<repo>
/plugin install <plugin>@<marketplace>

Claude Code prints the marketplace's name when it's added: that's the part after @. A git address, a local folder or a link to a marketplace.json file work too.

Anthropic's community marketplace, with plugins that outside authors submitted, needs adding too: /plugin marketplace add anthropics/claude-plugins-community, then install with @claude-community.

To pin a marketplace to a tag (a named release), add it after #: /plugin marketplace add your-org/plugins#v1.2.0. /plugin marketplace remove <name> removes one, along with every plugin you installed from it.

How to check a plugin before you install it

This part protects your business. I read every file of anything I install before it runs. A plugin is someone else's code running with your access: your code, your keys, any customer data on that computer. Claude Code's permission rules and its sandbox (the fence around Claude's own commands) limit Claude, not the plugin: its hooks and servers run with your full permissions, outside the sandbox. Even Claude Code's install warning says Anthropic doesn't control what's in a plugin.

1. Read what it runs. The details screen shows that a hook exists, not what it does. Choose Open homepage or View on GitHub and read hooks/hooks.json (each hook's command), .mcp.json (each server's command or address), every file in bin/ (tools Claude can run) and any script they call. In skills, look for allowed-tools (tools Claude may use without asking) and !`…` (a terminal command that runs when the skill is used). Ask Claude to explain any line you don't understand.
2. Prefer publishers you know. The name after @ names the catalog, not the author. Anthropic's marketplace names only work from Anthropic's own GitHub, so nobody can fake them, but most official plugins come from partners and other authors, and Anthropic doesn't review third-party marketplaces. claude plugin marketplace list shows where each one came from.
3. Pin or review updates. Auto-update can change the files you read in the background. It's on by default for Anthropic's official marketplace and off for most others, so check it for each one you add; switch it per marketplace on the Marketplaces tab of /plugin. For a marketplace you don't control, keep it off, pin it and update after reading what changed: claude plugin update <plugin>@<marketplace>.
4. Check what landed. claude plugin details <plugin> lists everything the plugin added and the tokens (the units your usage is counted in) it adds to every session.

Remove what you don't use

A forgotten plugin can still cost you usage, run servers and change under you. Clean up:

• The Installed tab in /plugin groups plugins you haven't used lately under Not used recently.
• Pause one with /plugin disable <plugin>. Remove it with /plugin uninstall <plugin>, or in your terminal with claude plugin uninstall <plugin> --scope <scope>, using the scope you installed it at.
• After an uninstall, its files stay in ~/.claude/plugins/cache/ for 14 days. If you no longer trust it, delete its folder there.
• Don't trust the marketplace's owner either? Remove the whole marketplace.

Next: write your own skill or set up hooks. Before launch: the production-ready checklist. The whole workflow: Claude Code best practices. All guides: articles.