Claude Code Plugins: What They Are and Which Ones Matter
Anthropic runs an official plugin directory, and on the day this was written it held 315 plugins from 147 different publishers. Most write-ups put the number near a hundred, which was true for a while and is not any more. The figures here come from reading the marketplace manifest directly rather than from another article.
The more useful thing to settle first is what a plugin actually is, because it is not the same as a skill and the two get used interchangeably.
A plugin is not a skill
A skill is one thing: a folder with a SKILL.md full of instructions. A plugin is a container that can hold several kinds of thing at once, and a skill is only one of them. Reading the first-party plugins in the official repository, this is what they actually contain:
| Component | What it adds |
|---|---|
skills/ | Instructions the model follows when the job matches. The most common component by far |
commands/ | Slash commands you type, namespaced as /plugin:command |
agents/ | Subagents with their own tools and prompt, for work you want delegated |
hooks/ | Code that fires on events, to block or reshape behaviour automatically |
scripts/ | Executables the plugin calls |
.claude-plugin/plugin.json | The manifest that declares everything above |
A plugin can also bring an MCP server with it, which is how it reaches an outside system. That is the practical dividing line. A skill tells the model how to do something. A plugin can additionally give it commands to run, agents to delegate to, hooks that fire without being asked, and a connection to your actual tools. If instructions are all you want, our list of Claude Code skills covers that side.
What is in the official directory
| Category | Plugins |
|---|---|
| Development | 123 |
| Productivity | 69 |
| Database | 39 |
| Monitoring | 22 |
| Security | 18 |
| Deployment | 9 |
| Design | 8 |
| Automation | 3 |
Anthropic publishes 40 of the 315. The rest come from companies integrating their own products: Google with 14, SAP with 9, Amazon Web Services with 7, then Shopify, Grafana, Oracle, Atlassian, Adobe, Auth0, CrowdStrike, Vanta and a long tail of one each. If you use a tool seriously, check whether its vendor has already shipped a plugin before you build anything yourself.
One caveat on that count. A marketplace entry is a listing, not a quality bar. Eighty-two of the 315 do not name an author at all, and most of the directory points at repositories owned by the publisher rather than code Anthropic reviewed. Treat it as a directory, which is what it calls itself.
The 40 Anthropic plugins, grouped
Worth knowing before you scan the list: twelve of the forty are language servers, one per language. They matter if you write that language and are irrelevant otherwise, so the number of general-purpose first-party plugins is closer to twenty-eight.
Language servers
typescript-lsp: TypeScript/JavaScript language server for enhanced code intelligencepyright-lsp: Python language server (Pyright) for type checking and code intelligencegopls-lsp: Go language server for code intelligence and refactoringrust-analyzer-lsp: Rust language server for code intelligence and analysisruby-lsp: Ruby language server for code intelligence and analysisphp-lsp: PHP language server (Intelephense) for code intelligenceswift-lsp: Swift language server (SourceKit-LSP) for code intelligencekotlin-lsp: Kotlin language server for code intelligencejdtls-lsp: Java language server (Eclipse JDT.LS) for code intelligencecsharp-lsp: C# language server for code intelligenceclangd-lsp: C/C++ language server (clangd) for code intelligencelua-lsp: Lua language server for code intelligence
Review and shipping
code-review: Automated code review for pull requests using multiple specialized agents with confidence-based scoring to filter false positivespr-review-toolkit: Comprehensive PR review agents specializing in comments, tests, error handling, type design, code quality, and code simplificationcommit-commands: Commands for git commit workflows including commit, push, and PR creationfeature-dev: Comprehensive feature development workflow with specialized agents for codebase exploration, architecture design, and quality reviewcode-simplifier: Agent that simplifies and refines code for clarity, consistency, and maintainability while preserving functionality. Focuses on recently modified code.code-modernization: Guided modernization for any legacy codebase: start with /modernize, get an assessment, an interactive map, the business rules mined from the code, and a plan…
Security
claude-security: Deep vulnerability scanning of your own code, run entirely inside your Claude Code session at a chosen effort tier, with every finding challenged before it is…security-guidance: Security review for Claude-generated code. Pattern-based warnings on edits, LLM-powered diff review on Stop, and an agentic commit reviewer that catches…
Building for Claude Code itself
plugin-dev: Comprehensive toolkit for developing Claude Code plugins. Includes 7 expert skills covering hooks, MCP integration, commands, agents, and best practices….skill-creator: Create new skills, improve existing skills, and measure skill performance. Use when users want to create a skill from scratch, update or optimize an existing…agent-sdk-dev: Development kit for working with the Claude Agent SDKmcp-server-dev: Skills for designing and building MCP servers that work seamlessly with Claude. Guides you through deployment models (remote HTTP, MCPB, local), tool design…mcp-apps: Skills for creating MCP Apps with the MCP Apps SDKmcp-tunnels: Connect Claude to a private MCP server through an Anthropic MCP tunnel. The /create-docker-mcp-tunnel command drives the Docker Compose quickstart end to end:…hookify: Easily create custom hooks to prevent unwanted behaviors by analyzing conversation patterns or from explicit instructions. Define rules via simple markdown files.claude-code-setup: Analyze codebases and recommend tailored Claude Code automations such as hooks, skills, MCP servers, and subagents.claude-md-management: Tools to maintain and improve CLAUDE.md files – audit quality, capture session learnings, and keep project memory current.
Reporting and output
session-report: Generate an explorable HTML report of Claude Code session usage , tokens, cache efficiency, subagents, skills, and the most expensive prompts , from local…receipts: A personal Claude Code impact report for justifying your usage to a manager or a self-review: what you shipped, which projects it went to, and each project’s…project-artifact: Generate and publish a living project status page , overview & success criteria, the workstream sequence, and next steps , as a shareable claude.ai artifact…explanatory-output-style: Adds educational insights about implementation choices and codebase patterns (mimics the deprecated Explanatory output style)learning-output-style: Interactive learning mode that requests meaningful code contributions at decision points (mimics the unshipped Learning output style)
Design and the rest
frontend-design: Create distinctive, production-grade frontend interfaces with high design quality. Generates creative, polished code that avoids generic AI aesthetics.playground: Creates interactive HTML playgrounds , self-contained single-file explorers with visual controls, live preview, and prompt output with copy button. Includes…ralph-loop: Interactive self-referential AI loops for iterative development, implementing the Ralph Wiggum technique. Claude works on the same task repeatedly, seeing its…math-proof: Two skills for hard mathematics problems, each ending in a self-contained proof.md: /math-proof:solo has the session work the problem itself in stages with a…math-olympiad: Solve competition math (IMO, Putnam, USAMO) with adversarial verification that catches what self-verification misses. Fresh-context verifiers attack proofs with…cwc-makers: Onboard a Code-with-Claude Makers Cardputer with one /maker-setup command , clones the build-with-claude repo, flashes UIFlow firmware, and installs the Claude…
Installing a plugin
The official directory is already available, so a plugin from it installs in one line inside Claude Code:
/plugin install code-review@claude-plugins-officialTo browse instead of guessing at names, /plugin on its own opens the
browser, and /plugin list shows what you already have.
Any other repository can be a marketplace. Add it once, then install from it by name:
/plugin marketplace add anthropics/skills
/plugin install document-skills@anthropic-agent-skillsThat second example is the skills repository, which is a separate marketplace from the plugin directory. Both are run by Anthropic and they are not the same catalogue, which is the most common reason an install command copied from an article does not work.
Read before you install
A plugin is not only instructions. It can carry hooks that run on events and scripts
that execute, so the honesty that applies to skills does not fully carry over: you can
read a SKILL.md in a minute, and you cannot audit a plugin that quickly.
Two habits worth keeping. Open the plugin’s repository and look at whether it ships
hooks/ or scripts/ before installing from a publisher you do
not recognise. And prefer the vendor’s own plugin for a tool you already pay for, since
the incentive to keep it working is clearer.
Building one
A plugin is a directory with a manifest. The minimum is
.claude-plugin/plugin.json plus at least one component, and
code-review is a good example of how small that can be: a manifest, a
licence, a readme, and a single command file.
my-plugin/
.claude-plugin/plugin.json
commands/my-command.md
skills/my-skill/SKILL.md
hooks/
agents/Anthropic ships plugin-dev for exactly this, described as a toolkit with
seven skills covering hooks, MCP integration, commands, agents and best practices. If you
are writing one, install that first rather than reverse engineering the format from
someone else’s repository.
To publish it, a marketplace is just a repository with a
.claude-plugin/marketplace.json listing your plugins. People add your repo
and install from it. There is no submission process in the way an app store has one.
Which ones to install first
The directory is large enough that installing broadly is a mistake. A plugin can add hooks that fire on every session, so a dozen of them interacting is genuinely hard to debug.
- The language server for your language. Dullest entry on the list and the one that changes the most. It gives the model real type information and definitions instead of inference from surrounding text.
- code-review, if you work on pull requests. It runs several specialised agents and scores findings by confidence specifically to cut the false positives that make automated review annoying.
- Your vendors’ plugins, for tools you already use daily. These are the ones that reach real systems, which is where plugins beat skills.
- plugin-dev or skill-creator, once you notice yourself repeating a prompt. That is the signal that the thing you want does not exist yet.
Then stop. Add one at a time and notice what changed, because the failure mode here is not a plugin that does nothing, it is three plugins quietly disagreeing.
What plugins do not do
They are not vetted software. The directory is Anthropic-managed, but most entries point at repositories owned by their publishers. Listing is not review.
They do not make the model better at your domain. A plugin changes what Claude can reach and what procedure it follows. It does not add judgement about your business, and the output still needs checking on anything that matters.
More is not better. This is the same trap as skills, with higher stakes, because plugins can execute. The useful set is small and matched to work you actually do.
Frequently asked questions
How many Claude Code plugins are there?
The official Anthropic-managed directory listed 315 plugins from 147 publishers when this was checked, of which Anthropic publishes 40. Figures near 100 that circulate in older articles were accurate earlier and are now out of date. Community registries index many more, but those are separate catalogues.
What is the difference between a Claude Code plugin and a skill?
A skill is a single folder with a SKILL.md containing instructions. A plugin is a container that can hold skills plus slash commands, subagents, hooks that fire on events, scripts, and an MCP server connection. So a skill tells the model how to do something, while a plugin can also give it commands, automation and access to outside systems.
How do I install a Claude Code plugin?
For the official directory, run /plugin install name@claude-plugins-official inside Claude Code. For any other repository, add it as a marketplace first with /plugin marketplace add owner/repo, then install from it. Running /plugin on its own opens a browser, and /plugin list shows what is already installed.
Are Claude Code plugins safe to install?
Treat the directory as a directory rather than a review process. Most entries point at repositories owned by their publishers, and 82 of the 315 do not name an author. Unlike a skill, which is readable Markdown with no code, a plugin can ship hooks and scripts that execute, so check whether a plugin from an unfamiliar publisher includes hooks or scripts before installing it.
Can I build my own Claude Code plugin?
Yes. A plugin is a directory with a .claude-plugin/plugin.json manifest plus at least one component, and some first-party plugins are as small as a manifest and a single command file. Anthropic publishes plugin-dev, a toolkit of seven skills covering hooks, MCP integration, commands and agents, which is the fastest way to get the format right.
How do I publish a plugin for other people to install?
Put a .claude-plugin/marketplace.json in a repository listing your plugins. Anyone can then add your repository with /plugin marketplace add and install from it. There is no submission or approval step like an app store, which is also why you should read what you install.
The short version: install the language server for whatever you write, add the plugins for tools you already pay for, and ignore the other three hundred until one of them solves a problem you actually have. If you want the instruction layer rather than the plumbing, start with the official skills.
