Claude Code Plugins

Plugins (Plugin) are, in Claude Code,the highest-level extension mechanism,used to take capabilities such as commands, agents, Skills, hooks, MCP, LSP andpackage, version, share, and distribute.。

A plugin = a reusable collection of Claude Code extension capabilities

A plugin can contain:

  • Slash Commands
  • Subagents (Agents)
  • Skills (capability descriptions)
  • Hooks (event hooks)
  • MCP servers (external tools/services)
  • LSP servers (code intelligence)

The core goal of plugins is only one:

Make Claude Code's capabilities reusable like a toolbox, rather than reconfiguring in every project.

Plugins vs Standalone Configuration (How to Choose)

Claude Code supports two extension methods:

Method Command form Suitable scenarios
Standalone configuration（.claude/） /hello Personal use, single project, quick experiments
Plugin（.claude-plugin/） /plugin-name:hello Team sharing, cross-project, versioned

When to use standalone configuration?

  • Only used in the current project
  • Personal workflow
  • Experimental configuration not yet stable
  • Want a short command name (such as/review）

When to use plugins?

  • Need toreuse in multiple projects
  • wantShare with team or community
  • Needversion control, upgrades, rollbacks
  • Plan to distribute through a marketplace
  • Can accept namespace commands (to avoid conflicts)

Best practices:

First, in.claude/the configuration, iterate → once stable, package as a plugin.

Minimal plugin structure (must remember)

my-plugin/
├── .claude-plugin/
│   └── plugin.json     # 插件清单（必需）
├── commands/           # 斜杠命令
├── agents/             # 子代理
├── skills/             # Skills
├── hooks/              # 钩子
├── .mcp.json           # MCP 配置
└── .lsp.json           # LSP 配置

Important rules

  • .claude-plugin/In the directory,can only containplugin.json
  • Other directories must be in the plugin root directory.

Plugin manifest (plugin.json)

The plugin's "ID card", determining:

  • Plugin name
  • Command namespace
  • Version
  • Author information

Example:

{
  "name": "my-first-plugin",
  "description": "A greeting plugin to learn the basics",
  "version": "1.0.0",
  "author": { "name": "Your Name" }
}

Key field descriptions:

Field Purpose
name Unique identifier + command namespace
description Displayed in the plugin marketplace
version Semantic versioning
author Optional, attribution description

Slash commands (most commonly used plugin capability)

1. Command definition method

  • Located incommands/directory
  • Each command = one Markdown file
  • Filename = command name

Example:

commands/hello.md

Corresponding command:

/my-first-plugin:hello

2. Command content example

---
description: Greet the user with a friendly message
---

Greet the user warmly and ask how you can help them today.

3. Command arguments

Usage$ARGUMENTSCapture user input:

Greet the user named "$ARGUMENTS" warmly.

Invoke:

/my-first-plugin:hello Alex

Local plugin testing (essential for development)

Usage--plugin-dirLoad plugin directory directly:

claude --plugin-dir ./my-plugin

Features:

  • No installation required
  • Restart Claude Code after modifications
  • Supports loading multiple plugins simultaneously
claude --plugin-dir ./plugin-a --plugin-dir ./plugin-b

What else can plugins do?

Capability Use case
Commands Custom slash commands
Agents Dedicated subagents
Skills Teach Claude when to use a certain capability
Hooks Automation (e.g., run commands after writing files)
MCP Connect to external services (GitHub, DB, API)
LSP Code intelligence (navigation, type checking)

Plugin Marketplace

Plugins are distributed throughthe marketplace; essentially, it is a plugin directory repository.

Official marketplace

  • Added by default
  • Run/plugin → Discover

Install plugin:

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

Plugin installation scope

Scope Description
User scope Only you, all projects
Project scope Current repository, shared with the team
Local scope Current repository, only you

Recommendation:

  • Team tools →Project scope
  • Personal productivity tools →User scope

Typical plugin categories

1. Code intelligence (LSP)

  • TypeScript, Python, Go, Rust, etc.
  • Provides definition navigation, references, and type errors

Requires local installation of the corresponding language server

2. External integration (MCP)

  • GitHub / GitLab
  • Jira / Notion
  • Slack / Figma
  • Vercel / Supabase

Essence:Plugin = MCP server + configuration

3. Development workflow

  • Git commits, PRs
  • Code review agent
  • Plugin development tools

Common commands for plugin management

/plugin                # 打开插件管理器
/plugin install         # 安装插件
/plugin uninstall       # 卸载
/plugin enable/disable  # 启用 / 禁用
/plugin marketplace add # 添加市场
/plugin marketplace rm  # 移除市场

from.claude/Migrating to plugins (core ideas)

Before After migration
.claude/commands plugin/commands
.claude/agents plugin/agents
settings.json hooks plugin/hooks/hooks.json

After migration:

  • Plugin version takes effect first
  • Can delete old.claude/Configuration avoids duplication

When must you definitely use plugins?

  • You already havestable Claude workflows
  • You arerepeatedly copying.claude/
  • Team members start asking you: "How do I configure this?"
  • You want Claude to be as controllable as IDE plugins

Plugins are the watershed moment for Claude Code transitioning from a "personal AI assistant" to an "engineering-grade tool"

Other extensions