A Claude Code plugin is a folder that brings several extensions at once: skills, subagents, hooks, MCP servers and other components that Claude Code installs and loads as one unit. You install plugins in a session with /plugin, pick them from a marketplace on the "Discover" tab, and decide whether they apply to you, to the whole repository, or only locally. Claude Code adds Anthropic's official marketplace by itself the first time you start an interactive terminal session; you add other catalogs with /plugin marketplace add. One thing first: according to Anthropic, a plugin can run arbitrary code with your user privileges. Install only what you have checked, from a source you trust. At its core, a plugin of your own is a plugin.json plus folders for its components; since September 25, 2026, Anthropic has accepted plugins for its directory through a developer portal.
I work with Claude Code every day and run my team of fourteen AI employees on it. I use plugins from the official catalog there too. Everything about Claude is collected on my Claude page; if you have not set up Claude Code yet, start with Claude Code: A Guide for Beginners.
What a Claude Code plugin is, and when you don't need one
Anthropic describes a plugin as "a directory of skills, agents, hooks, MCP servers, or other components that Claude Code installs and loads as one unit". You can use each of these components without a plugin. A skill in ~/.claude/skills/, for example, already applies to every project on your machine. The docs draw the line clearly: a plugin pays off when you want to share several components as a package, install them in many projects, or publish versioned releases.
An enabled plugin is part of every session, not only the ones where you use it. For each skill and agent that Claude may invoke on its own, the name and description sit in the context on every turn. That costs usage and room in the context window even when nothing from the plugin runs. That is why I advise using plugins sparingly.
The skills of my AI workforce live in one shared Git repository and reach every employee through Git. A plugin would be the next step if I wanted to ship the same components, versioned, to many other projects or to other teams.
Marketplaces: official, community and third-party
According to the docs, a marketplace is "a catalog, not a hosted store": a repository with a .claude-plugin/marketplace.json file that lists plugins and says where Claude Code fetches each one. Anthropic runs three general catalogs:
Anthropic also publishes topic-specific catalogs such as anthropics/skills. For plugins listed in both the demo and the official catalog, such as code-review or commit-commands, the docs say to install from the official one so you do not end up with two copies.
A marketplace's name tells you who publishes the catalog, not what a single plugin in it does. Claude Code accepts the official and community names only for catalogs from github.com/anthropics/, so a third-party catalog cannot pass itself off as Anthropic's. All others are third-party, including a catalog from a colleague or your own company. Many well-known plugins only live in such catalogs, usually the author's own GitHub repository. Anthropic does not review those catalogs, according to the docs.
Two things get mixed up a lot. Claude Marketplace at claude.com/marketplace is a website for browsing, not a catalog for /plugin marketplace add. And Anthropic's directory is the catalog on claude.ai, from which plugins reach Claude Code through account sync, where they show up as <name>@synced.
How to install and manage Claude Code plugins
In the terminal, installing takes five steps. The docs use commit-commands from the official catalog as the example, which adds commands for committing, pushing and pull requests:
/plugin install commit-commands@claude-plugins-official
- Open the details. In a session this command does not install right away; it opens the plugin's details in the
/plugin panel. Without a name, /plugin opens the "Discover" tab with every plugin from your catalogs.
- Check what comes with it. "Will install" lists commands, agents, skills, hooks and servers. For plugins from the official catalog, the view also shows "Context cost": what the plugin adds to every message and what it costs when invoked.
- Choose a scope. User (you, in every project, entry in
~/.claude/settings.json), Project (everyone in the repository, entry in the committed .claude/settings.json) or Local (only you, in this repository).
- Read the summary. It tells you whether the plugin is active now or needs
/reload-plugins.
- Test it. Type
/ and look for /commit-commands:commit.
Without a session, the same works from your shell, for example in a setup script:
claude plugin install commit-commands@claude-plugins-official
claude plugin list
claude plugin disable commit-commands@claude-plugins-official
claude plugin details commit-commands
In claude plugin details, the "Always-on" line shows how many tokens the plugin adds to every session. On the "Installed" tab, Claude Code groups plugins you have not used for a while under "Not used recently". Switch off whatever you do not use.
For the official catalog, Claude Code fetches updates automatically by default; for community and third-party catalogs, auto-update is off. Two limits are worth knowing. Cloud sessions, including the browser at claude.ai/code, do not load the plugins on your machine. And in a claude -p run, /plugin does not work, although plugins you already installed do load.
The best Claude plugins? Start with the official catalog, not a ranking
Anthropic does not publish a ranking of the "best Claude Code plugins", and I am not going to make one up. According to the docs, the official catalog changes often; when I fetched it on October 26, 2026, its catalog file had 315 entries, most of them from partners and other authors. Anthropic maintains a smaller set of its own. These are a good starting point because they come from the official catalog and update automatically:
How the code-review plugin differs from Anthropic's Code Review service and from the /code-review command is covered in Claude Code Review: Cost, Setup, Limits.
My rule of thumb: one plugin per concrete problem you have today.
Security: plugins run code with your privileges
The first sentence of Anthropic's security page reads: "A Claude Code plugin you install can execute arbitrary code on your machine with your user privileges." A plugin's hooks and monitors run as shell commands, MCP and language servers as their own processes, and executables in its bin/ folder land on the Bash tool's path. According to the docs, your permission rules and the sandbox cover Claude's tool calls, not the code a plugin starts by itself.
In the details pane before you install, Claude Code shows the same warning, whatever the catalog: "Make sure you trust a plugin before installing, updating, or using it. Anthropic does not control what MCP servers, files, or other software are included in plugins and cannot verify that they will work as intended or that they won't change." The docs list four checks, and I recommend all of them:
- Check the catalog's source with
claude plugin marketplace list.
- Read the details view: what is under "Will install"?
- Read the source:
hooks/hooks.json (which command runs), .mcp.json (which server, which address) and every file in bin/.
- Print the inventory:
claude --plugin-dir <folder> plugin details <name> lists every component without starting a session.
One example of why step 1 matters: on October 26, 2026, Anthropic's community catalog listed several plugins by different authors whose names echo that of a well-known plugin from the official catalog. A similar name says nothing about the author. Watch auto-update as well: if it is on for a catalog, the files you reviewed can change in the background. Companies can allow or block catalogs and require plugins through managed settings; according to the release notes of August 6, 2026, Enterprise admins can also turn on a beta security scan that automatically checks third-party skills and plugins for malicious content when someone uploads or edits them. What else belongs to securing Claude Code is covered in my piece on Claude Code security.
Mods: plugins that change Claude Code itself
Mods are described in the official docs. According to Anthropic, a mod is "a plugin that changes how Claude Code looks and behaves": a plugin with a hooks module in JavaScript or TypeScript that runs inside Claude Code. A mod can draw its own panes beside the transcript, restyle parts of the interface, hold or rewrite a tool call, and add commands that run instantly. Mods work in the terminal, in the Code tab of the desktop app and in the VS Code extension; in the terminal from version 2.1.287.
For security, mods call for even more care than other plugins. The docs are blunt: "Mods aren't sandboxed." A mod can read and write files, read environment variables including API keys, see every prompt and every tool call, and approve a call before you are asked. With a Team or Enterprise sign-in or managed settings, that approval does not get past your ask rules and PreToolUse hooks by default. A mod cannot restyle the permission prompt itself. Before you install a mod, claude plugin validate ./folder shows in its hooks: and calls: lines what it reacts to and what it requests. For a single session, --safe-mode turns all mods off, along with your other customizations.
Building your own plugin: the skeleton
You do not need a marketplace to test a plugin. The smallest useful example in the docs is a plugin with a single skill:
mkdir -p my-first-plugin/.claude-plugin
mkdir -p my-first-plugin/skills/hello
# create my-first-plugin/.claude-plugin/plugin.json (see below)
# create my-first-plugin/skills/hello/SKILL.md
claude plugin validate ./my-first-plugin
claude --plugin-dir ./my-first-plugin
{
"name": "my-first-plugin",
"description": "A greeting plugin to learn the basics",
"version": "1.0.0",
"author": { "name": "Your Name" }
}
Only name is required; it becomes the prefix of every skill and agent, so the skill runs as /my-first-plugin:hello. Only plugin.json goes into the .claude-plugin/ folder; every component sits next to it: skills/, agents/, hooks/hooks.json, .mcp.json. claude plugin validate ends with "Validation passed" or names the field that is wrong. With --plugin-dir you load the plugin for exactly one session without installing it.
For larger plugins, Anthropic's plugin-dev plugin helps: with /plugin-dev:create-plugin and a description, Claude plans, builds and checks it with you. If your skills, agents and hooks already live under .claude/, the docs describe how to move them into a plugin; /hello then becomes /<plugin>:hello.
Publishing: your own marketplace or Anthropic's directory
There are three ways to share a plugin. Send the folder or a .zip directly, in which case there are no updates. Set up your own marketplace: a .claude-plugin/marketplace.json in the repository is enough, and a private repository makes a private catalog. Or submit it to Anthropic: since September 25, 2026, Anthropic has accepted plugins for its directory through a developer portal at claude.ai/directory/manage, with review status and usage analytics once they are live.
To submit, you need a paid Claude plan and a GitHub repository. On Pro and Max you submit yourself; on Team and Enterprise, an Owner does. Every version goes through automated validation and a security scan, and a person reviews a new listing before it goes live; Anthropic gives no fixed review time. The official catalog claude-plugins-official does not take submissions through this portal.
Frequently asked questions
What is the difference between a plugin and a skill?
A skill is a single set of instructions in a SKILL.md. A plugin is a package that can hold one or more skills, plus agents, hooks, MCP servers and a hooks module. For a single skill that only you use, you do not need a plugin.
How do I install a Claude Code plugin?
In a session with /plugin install <name>@<marketplace>, or through the "Discover" tab in /plugin. If the plugin is not in the official catalog, add its marketplace first with /plugin marketplace add <owner>/<repo>. From your shell, use claude plugin install.
Are Claude Code plugins safe?
Not by themselves. According to Anthropic, a plugin can run arbitrary code with your privileges, and Anthropic does not review third-party catalogs. Go through the four checks above before you install.
What are Claude Code mods?
Plugins with a hooks module in JavaScript or TypeScript that runs inside Claude Code and changes its interface and behavior. According to the docs, mods run without a sandbox, with your privileges.
Where to go from here
Start with a single plugin from the official catalog and look at its context cost before you install it. If you then want to bundle your own workflows, first build a skill and a hook without a plugin, as described in my article on Claude skills and in Claude Code Hooks: Rules That Always Apply, and only package them as a plugin once you want to share them. How others build their plugins and AI employees is what we discuss in my German-speaking community Claude Practitioners.
