← claude-core
Resources
▮ CLAUDE-CORE · ACADEMY CLAUDE CODE v2.1.143 14 PLUGINS · 6 CATEGORIES ▮

PLUGIN MARKETPLACE

Most Claude Code users still run vanilla. The real power lives in the official plugin marketplace — preinstalled, /plugin away, and shaped around the same primitives this site teaches (hooks, skills, MCP, subagents). This page covers the discovery UX, the canonical entry point (claude-code-setup), 14 hand-picked plugins across 6 categories, and the security model you should hold before installing anything.

What a marketplace is, in 30 seconds

A marketplace is a catalog of plugins. A plugin is a bundle of skills + agents + hooks + MCP servers + slash commands + LSP integrations — same primitives you'd write by hand, packaged for distribution. The official Anthropic marketplace (claude-plugins-official) is auto-registered when you start Claude Code. Other marketplaces — including Anthropic's own demo, your team's private one, or any third-party — you add explicitly.

The two-step model

  1. Add the marketplace — registers the catalog. No plugins installed yet.
  2. Install individual plugins — browse, pick, install at one of three scopes.

App store metaphor: adding the store gives you the browse surface; install is per-app, per-scope. The whole flow lives behind one slash command — /plugin.

§1

The /plugin command — four tabs

What/plugin opens a four-tab manager. Tab through with Tab (or Shift+Tab backward).

▸ CLI shortcuts (skip the UI)

# Install (defaults to User scope)
/plugin install <plugin>@<marketplace>

# Manage
/plugin disable <plugin>@<marketplace>
/plugin enable  <plugin>@<marketplace>
/plugin uninstall <plugin>@<marketplace>

# Pick up changes mid-session, no restart
/reload-plugins

# Shortcuts: /plugin market === /plugin marketplace; rm === remove

Gotcha/reload-plugins is the most-forgotten command. If you install a plugin and its skills don't show up, you almost always just needed to reload. After reload, the output shows counts for plugins, skills, agents, hooks, MCP servers, and LSP servers — confirm the new plugin's contributions appear.

§2

Three install scopes — pick deliberately

ScopeWhoWhere it livesUse when…
user (default) You, every project ~/.claude/settings.json You'd use this everywhere — commit-commands, github, an LSP for your daily language.
project You + every collaborator on this repo .claude/settings.json (git-tracked) The plugin is part of the project's contract — a code-review agent the whole team should run, a hook policy.
local You, this repo only .claude/settings.local.json (gitignored) Experimenting; not ready to bless for the team; your personal observability stack.
managed Org admin pushed it Managed settings file Read-only from your side. You see it; you can't disable it.

▸ CLI scope flag (bypasses the picker)

claude plugin install formatter@your-org --scope project
claude plugin uninstall formatter@your-org --scope project

Gotcha — installing at user scope (the default) means the plugin is active in every repo you open. If the plugin adds an aggressive hook or output style, it bleeds into projects where you didn't want it. When in doubt, install at local first, validate, then promote.

§3

The canonical entry point — claude-code-setup

Why — staring at a marketplace of 200+ plugins isn't where you want to start. claude-code-setup is a single read-only skill, published by Anthropic, that scans your codebase and recommends the top 1-2 automations per category. It's the bootstrapping primitive for any existing repo adopting Claude Code.

▸ Install

/plugin install claude-code-setup@claude-plugins-official
/reload-plugins

▸ Use — any of these prompts triggers it

"recommend automations for this project"
"help me set up Claude Code"
"what hooks should I use?"

▸ What it recommends, by category

Safety — the skill is read-only. It analyzes, it doesn't install. After the recommendation lands, you decide which plugins to install and at what scope. Treat the output as a starting point for /plugin install, not as ground truth — the recommendation reflects what the skill saw, not necessarily what your project needs long-term.

§4

14 plugins worth knowing — curated

How to read this — the marketplace ships 200+ plugins and moves fast. The 14 below were picked for breadth (one or two per category) and staying power (built or sponsored by stable upstream maintainers). For the live catalog, run /plugin and tab to Discover, or visit claude.com/plugins.

#1

claude-code-setup

bootstrap

Read-only analyzer that scans your codebase and recommends the top 1-2 automations per category (MCP / Skills / Hooks / Subagents / Slash commands).

/plugin install claude-code-setup@claude-plugins-official
#2

claude-md-management

bootstrap

Maintain and improve CLAUDE.md files — the project-orientation file every Claude Code session reads first.

/plugin install claude-md-management@claude-plugins-official
#3

github

external integrations

Official GitHub MCP server — repos, issues, PRs, releases, code search, file edits via gh-API auth.

/plugin install github@claude-plugins-official
#4

gitlab

external integrations

GitLab DevOps integration — merge requests, pipelines, CI variables, project search.

/plugin install gitlab@claude-plugins-official
#5

atlassian

external integrations

Jira + Confluence — read tickets, comment, transition status, search docs.

/plugin install atlassian@claude-plugins-official
#6

context7

external integrations

Upstash docs lookup MCP — fetches up-to-date library documentation by name instead of hallucinated APIs.

/plugin install context7@claude-plugins-official
#7

commit-commands

development workflows

Slash-commands for the git commit + push + PR flow — drafts the message, stages, commits, pushes.

/plugin install commit-commands@claude-plugins-official
#8

code-review

development workflows

Specialized agent that reviews a PR or branch — security, performance, accessibility, naming sub-agents in one dispatch.

/plugin install code-review@claude-plugins-official
#9

frontend-design

development workflows

Skill for distinctive, production-grade frontend UI — moodboarding, design tokens, component composition.

/plugin install frontend-design@claude-plugins-official
#10

typescript-lsp

code intelligence (LSP)

TypeScript language server — automatic diagnostics on every edit, jump-to-def, find-refs, hover types.

/plugin install typescript-lsp@claude-plugins-official
#11

pyright-lsp

code intelligence (LSP)

Pyright (Microsoft) language server — Python type checking + diagnostics in the same turn as the edit.

/plugin install pyright-lsp@claude-plugins-official
#12

rust-analyzer-lsp

code intelligence (LSP)

rust-analyzer language server — Rust type errors, borrow checker, trait resolution surfaced inline.

/plugin install rust-analyzer-lsp@claude-plugins-official
#13

explanatory-output-style

output styles

Output style that prepends an educational note about WHY a given implementation choice was made — useful in pairing/onboarding.

/plugin install explanatory-output-style@claude-plugins-official
#14

claude-code (demo marketplace)

third-party

Anthropic's reference plugins, not preinstalled — adds via /plugin marketplace add anthropics/claude-code. Useful as templates when authoring your own.

/plugin install <plugin>@anthropics-claude-code

Gotcha — LSP plugins require the language server binary to be on your $PATH (e.g., typescript-language-server, pyright-langserver, rust-analyzer). If the /plugin Errors tab shows Executable not found in $PATH, the plugin installed correctly — you just need to install its sibling binary via your package manager (brew, npm, cargo, system package).

§5

Adding third-party marketplaces

When — your team has a private plugin catalog, you want to try a community marketplace, or you're testing a marketplace you're building. The official one ships preinstalled; everything else needs an explicit add.

▸ From a GitHub repo (most common)

/plugin marketplace add owner/repo

# Anthropic's demo marketplace (separate from the official)
/plugin marketplace add anthropics/claude-code

▸ From any Git host (include .git suffix)

/plugin marketplace add https://gitlab.com/company/plugins.git
/plugin marketplace add git@gitlab.com:company/plugins.git

# Pin to a specific branch or tag
/plugin marketplace add https://gitlab.com/company/plugins.git#v1.0.0

▸ From a local path or remote URL

/plugin marketplace add ./my-marketplace
/plugin marketplace add ./path/to/marketplace.json
/plugin marketplace add https://example.com/marketplace.json

▸ Manage the registered set

/plugin marketplace list
/plugin marketplace update <name>
/plugin marketplace remove <name>     # uninstalls every plugin from it!

Gotchamarketplace remove is destructive. Removing a marketplace uninstalls every plugin you sourced from it. If you only want to stop auto-updates, disable the auto-update toggle in the Marketplaces tab instead.

§6

Team marketplaces — auto-prompt on repo trust

When — the team has a shared plugin catalog and you want new contributors to install it without onboarding friction. Declared in .claude/settings.json, prompted on first repo trust.

.claude/settings.json

{
  "extraKnownMarketplaces": {
    "my-team-tools": {
      "source": {
        "source": "github",
        "repo": "your-org/claude-plugins"
      }
    }
  }
}

When a teammate first opens the repo and trusts it, Claude Code prompts them to install the listed marketplaces and any enabledPlugins declared alongside. They opt in or skip; the repo doesn't enforce.

Gotcha — third-party and local-development marketplaces ship with auto-update disabled by default (only official Anthropic marketplaces auto-update). Admins can flip "autoUpdate": true per-entry in managed settings if the org wants forced refreshes. For most teams, manual /plugin marketplace update at known stable points is healthier than silent updates.

§7

The security model — earn the trust

What's at stake — plugins and marketplaces execute arbitrary code with your user privileges. A plugin can run shell, read your files, hit the network, modify your environment. Anthropic does not audit plugin code; the claude-plugins-official badge means Anthropic curates the catalog, not that the runtime is sandboxed.

▸ The two-line audit reflex

# 1. Where does the plugin live? (read this before installing)
gh repo view <owner>/<repo> --json url,description,homepageUrl

# 2. What hooks / commands / MCP does it ship? (look inside .claude-plugin/)
gh repo clone <owner>/<repo> && grep -RIn 'hook\|command\|mcpServers' .claude-plugin/

Gotcha — official marketplaces auto-update by default. Even the official ones can ship a bad release, so if a session breaks after a startup notification ("plugins updated"), check /plugin → Marketplaces → toggle off auto-update for the suspect entry, then bisect. FORCE_AUTOUPDATE_PLUGINS=1 + DISABLE_AUTOUPDATER=1 keeps plugin updates while pinning the CLI itself — useful when you've stabilized a workflow.

§8

Troubleshooting — the four real failure modes

Where this slots into the rest of claude-core

The marketplace is the discovery layer. The other surfaces on this site teach the primitives that plugins are built from:

Plugins are how the ecosystem ships expertise; primitives are how you read what they did and customize it. Both directions matter.