The best Claude Code plugins, and what each one runs

How we picked

We ranked plugins by installs on claude.com/plugins (opens in a new tab), Anthropic's plugin directory, on November 7, 2026, and grouped them by what they're for. For each one we read its source at the version Anthropic's official marketplace (claude-plugins-official) points to, and noted what it adds to Claude Code. We didn't score them on taste: what a plugin runs is a fact you can check.

What a plugin can run on your machine

A plugin is a bundle of components. Some are text for Claude, some run code, and Claude Code's docs put it plainly: what the plugin runs, it runs as you.

• Skills, commands and agents are instructions. Claude reads them when they're relevant or when you run them; nothing runs on its own.
• Hooks are commands Claude Code runs automatically at points in a session: when it starts, before or after an edit or a shell command, when Claude stops.
• MCP servers are tool servers. A local one is a process started on your machine in every session where the plugin is enabled. A remote one is a URL, and what Claude sends it goes to that host.
• Language servers (LSP) run in the background to give Claude code intelligence for a language.

An enabled plugin is part of every session, not only the ones where you use it: its skill names and descriptions sit in Claude's context on each turn, its hooks fire at their events and its MCP servers run.

The most-installed Claude Code plugins at a glance

Installs from claude.com/plugins and components from each plugin's source, both checked on November 7, 2026.

Plugins for how you build

Superpowers is the most-installed community plugin (MIT). Its 15 skills teach Claude to brainstorm, plan, develop with subagents, debug systematically and work test-first. Its one hook runs at session start and loads its introductory skill into the session, so Claude knows the others exist.

Feature Dev, Code Review and Code Simplifier come from Anthropic: a guided feature workflow with agents for exploring the codebase, designing and reviewing; a pull request review by several agents that scores each finding to filter out false positives; and an agent that tidies recently changed code. All three are instructions only.

Commit Commands adds commands to commit, push and open a pull request, and Skill Creator helps you write, test and benchmark your own skills. Neither runs anything on its own.

Plugins for docs, browsers and services

Context7, from Upstash, pulls current, version-specific documentation and code examples into Claude's context. It's a remote MCP server: nothing installs locally, and your documentation lookups go to Context7's servers.

Playwright, Microsoft's browser automation server, lets Claude open pages, click, fill in forms and take screenshots, which makes it the usual pick for end-to-end tests. It starts locally with npx @playwright/mcp@latest, so each session runs whatever version npm currently marks as latest. That keeps it current; it also means a bad release published to npm could run on your machine at your next session.

GitHub's official MCP server lets Claude create issues, manage pull requests and search code through GitHub's API. It's remote, at api.githubcopilot.com.

Frontend Design, the most-installed plugin of all, is a single skill that steers Claude away from generic-looking interfaces. TypeScript LSP adds a language server for TypeScript and JavaScript; the directory has similar ones for Python, Go, Rust and other languages.

Claude Code security plugins

Security plugins sit closest to your code, and most of them run the most. In order of installs:

• Security Guidance (Anthropic, 241,800 installs) warns about risky patterns as Claude edits, reviews the diff with an LLM when Claude stops, and adds a commit reviewer for injection, XSS, SSRF, hardcoded secrets and 25+ other vulnerability classes. Its hooks run at session start, on every prompt, after every edit and shell command, and at stop. The session-start hook installs Anthropic's Agent SDK into a virtual environment under ~/.claude/security/ if Python can't already import it.
• Semgrep (19,548) catches vulnerabilities as Claude writes code and steers it toward secure patterns. It runs hooks before and after edits and shell commands, at session start and end and when Claude stops, plus a local MCP server.
• Claude Security (Anthropic, 8,580) runs a deep vulnerability scan of your own code inside your session, challenges every finding before it reports it, and proposes patches you apply when you choose. It adds 9 agents and hooks that run after shell commands.
• Sonatype Guide (7,484) checks dependencies for vulnerabilities and recommends secure versions, through Sonatype's remote MCP server.
• Aikido (7,322) brings Aikido's SAST, secrets and IaC scanning through a local MCP server pinned to an exact version (@aikidosec/mcp@1.0.28).
• Pharon (ours, from our own marketplace) answers a different question: when a package, GitHub Action or vendor you use is compromised, did this machine install it, and what should you rotate? Its hooks run pharon sync when a session starts and ends to refresh your stack's inventory, names and versions only, and do nothing until you've confirmed the first upload. It also shows open alerts in the status line. Apache-2.0.

The scanners check the code you and Claude write. None of them watches what's already installed when an incident lands; the latest supply chain notices show what that looks like.

How to check a plugin before you install it

1. Look at the marketplace. Claude Code accepts the names of Anthropic's official and community marketplaces only from repositories under github.com/anthropics; every other marketplace is third-party. Whatever the tier, a plugin runs with your user privileges.
2. Open the plugin in `/plugin`. The Marketplaces tab shows its details, and plugins in Anthropic's official marketplace show a context cost estimate before you install.
3. Read its hooks and servers. hooks/hooks.json lists what runs automatically and when; .mcp.json lists the servers it starts or the hosts it talks to. A few lines of JSON tell you more than the README.
4. Prefer pinned versions. A server started with @latest runs whatever is newest on the registry at your next session; an exact version only changes when the plugin does.
5. Turn off what you don't use. Disable a plugin in /plugin, or with claude plugin disable in your shell, without uninstalling it.

Plugins are a software supply chain like any other. In October 2026 a tampered Claude Code extension sat in Cursor's extension marketplace for about 11 hours and installed an SSH backdoor. It was an editor extension, not a Claude Code plugin, but the lesson carries over: Pharon's notice has the details.

How to install a Claude Code plugin

Run /plugin in Claude Code and pick one from the Discover tab, which lists the plugins in your marketplaces. Claude Code adds Anthropic's official marketplace the first time you start it, so you can also install by name:

/plugin install commit-commands@claude-plugins-official

For a plugin from another marketplace, add the marketplace first. Pharon's, for example:

/plugin marketplace add getpharon/pharon
/plugin install pharon@pharon

Sources

• Claude Code plugins overview (opens in a new tab), Plugin security and trust (opens in a new tab)
• claude.com/plugins (opens in a new tab) for installs, Anthropic's official marketplace (opens in a new tab) for each plugin's source
• Superpowers (opens in a new tab), Semgrep plugin (opens in a new tab), Aikido plugin (opens in a new tab), Sonatype Guide plugin (opens in a new tab), Pharon plugin (opens in a new tab)