Claude Code Plugins and Marketplaces: How to Install, Top Picks, and Extending Your Setup (2026)
A Claude Code plugin is a prepackaged bundle of extensions - skills, subagents, hooks, slash commands, and MCP servers - that you install once and use right away. A marketplace is the catalog you browse to find and install those plugins. Open it with the /plugin command, go to the Discover tab, then run /plugin install <name>@claude-plugins-official. This guide covers what a plugin actually contains, how to install one, which plugins are worth installing (with notes on context cost), and when a paid all-in-one kit makes more sense than hand-collecting free plugins.
the Claude Code plugin system moves fast, so tab names and commands may differ slightly between versions. The commands below are checked against the official Claude Code documentation (accessed 08/09/2026).
What are Claude Code plugins and marketplaces?
Claude Code's plugin system shipped in late 2025, and it is the cleanest way to bolt extra capabilities onto the tool without configuring every piece by hand.
A plugin is a prepackaged extension that bundles several components - skills, subagents/agents, hooks, slash commands, MCP servers, and even an LSP server - so you install it once instead of wiring each piece up manually. For example, rather than writing your own slash command to generate a commit, configuring a hook to run before a save, and then declaring an MCP server for GitHub, you install a single plugin that packages all of that together.
A marketplace is the catalog that holds plugins for you to browse and install. Think of it this way: a plugin is the "app," and a marketplace is the "app store." Under the hood, a marketplace is just a Git repo (or a folder, or a URL) with a manifest file that lists the plugins it offers. Anthropic runs the official marketplace, the community has its own, and you can spin up an internal marketplace for your team too.
Before the plugin system existed, adding a capability meant creating a skill file yourself, declaring an MCP server in your config, and writing a hook by hand - each in a different place, and every time you switched machines or shared with a teammate, you started over. A plugin wraps all of that into one tidy, versioned unit you can install, remove, and share through a marketplace. That is the big shift: from "scattered config" to "install it like an app."
The nice part: everything is a plugin under the same standard, so once you understand how one plugin works, you understand them all. If you are still fuzzy on the building blocks, read what Claude Code skills are and the breakdown of skills vs subagents vs hooks vs MCP first - those two make the next section much easier to follow.
What components does a plugin contain?
This is the part that trips up newcomers the most, because the concepts all sound alike. A plugin can hold any combination of the six component types below - it does not have to include all six.
| Component | What it is | When to use it |
|---|---|---|
| Skill | A packaged set of instructions plus resources that Claude loads on its own when it hits the right kind of task. | You want Claude to reliably handle a repeating task (writing tests, processing PDFs, etc.). |
| Subagent / agent | A "sub-assistant" with its own prompt and context, running separately from the main session. | You need to split a big job into independent tasks and keep the main session's context lean. |
| Hook | A snippet that fires automatically at an event (before/after a tool call, on file save, etc.). | You need to enforce a rule: run a linter, block a dangerous command, auto-format. |
| Slash command | A quick /name command that triggers a predefined workflow. | You want fast access to a task you run often (create a commit, review a PR). |
| MCP server | A bridge that lets Claude talk to an outside service (GitHub, Notion, a database, etc.). | You need Claude to read from or write to an external system. |
| LSP server | A Language Server that gives deep, language-specific understanding (definitions, type errors, etc.). | You want Claude to "understand" code in a specific language (Python, TypeScript, etc.). |
A quick way to remember it: a skill teaches Claude what to do, a subagent is who does it, a hook is when it runs automatically, a slash command is how you call it, and MCP and LSP are the connections to the outside. To go deeper on each type, see the subagents guide, Claude Code hooks, what MCP is, and slash commands; and if you want to package a capability of your own, there is a walkthrough on creating a custom skill.
How to install a plugin in Claude Code
Installing a plugin always follows a two-step model: (1) add a marketplace, then (2) install a plugin from that marketplace. For the official marketplace, step 1 is basically automatic - Claude Code adds it on startup, so you usually only need step 2.
The most visual way is the TUI. Type this in a Claude Code session:
/pluginThe /plugin screen has four tabs:
- Discover - browse installable plugins from the marketplaces you have added.
- Installed - the plugins you have; enable/disable, uninstall, mark as favorite.
- Marketplaces - manage the marketplace sources you have added.
- Errors - reports when a plugin fails to load.
In the Discover tab, pick the plugin you want, choose a scope (User / Project / Local - explained in the management section), then confirm. If you prefer typing commands, you can install directly without opening the TUI:
/plugin install github@claude-plugins-officialThe syntax is <plugin-name>@<marketplace-name>. To install outside a session (say, in a setup script), use the CLI:
claude plugin install github@claude-plugins-official --scope projectOn scope, the three common choices when installing are: User (installed for your account, available in every project), Project (this project only, and you can commit it to Git so the whole team shares it), and Local (this machine only, not shared). My rule of thumb: plugins I use everywhere (git, review) go to User scope; plugins tied to one project (an internal MCP, a language-specific LSP) go to Project scope.
After installing, run /reload-plugins to activate it without restarting the session. One easy gotcha: slash commands from a plugin are usually namespaced under the plugin name to avoid collisions - for example, the commit command from the commit-commands plugin is called /commit-commands:commit, not /commit. If a plugin does not show up after install, open the Errors tab in /plugin - it is usually a syntax error in a config file or a marketplace that was not added correctly.
Marketplaces: official, community, and adding your own
Anthropic provides three marketplace sources:
claude-plugins-official- the official marketplace, added for you on Claude Code startup. You can browse this catalog on the web at claude.com/plugins (accessed 08/09/2026).claude-plugins-community- community plugins, added manually with:
/plugin marketplace add anthropics/claude-plugins-community- The demo marketplace
anthropics/claude-code- a few illustrative examples to study the structure.
You can also add your own marketplace from several kinds of source: a GitHub repo in owner/repo form, a git URL, a local folder path, or a remote URL. The general syntax is /plugin marketplace add <source>.
For teams, the important trick is to declare it in .claude/settings.json:
{
"extraKnownMarketplaces": {
"team-tools": {
"source": { "source": "github", "repo": "your-company/claude-plugins" }
}
}
}Commit that file to Git and the whole team shares one plugin source - anyone who clones the repo sees the exact same set of marketplaces, with no need to remind each other to add it by hand. To list the marketplaces you have, use /plugin marketplace list; to drop a source you no longer need, use /plugin marketplace remove <name>. This is also how a team keeps its source list tidy and avoids the case where everyone adds a different source and accidentally installs a name-clashing plugin from two marketplaces.
To give you a sense of what the official marketplace holds, here are the common plugin groups:
| Group | Example plugins |
|---|---|
| Code intelligence (LSP) | Language-specific LSPs: pyright-lsp, typescript-lsp, etc. |
| External integrations (MCP) | github, gitlab, atlassian, linear, notion, figma, vercel, supabase, slack, sentry |
| Security | security-guidance |
| Dev workflows | commit-commands, pr-review-toolkit, plugin-dev |
| Output styles | Plugins that change the response style or output format |
Top Claude Code plugins worth installing (2026)
Here is an honest, curated list - I picked these on a "actually use it, worth the context" basis. Do not install them all at once (see the context-cost note at the end of this section).
- commit-commands - standardizes commits and PR creation. Worth it if you commit several times a day and want consistent messages.
- pr-review-toolkit - pull request review with a checklist. Worth it if you or your team review PRs regularly.
- security-guidance - flags and reviews security holes as you code. Worth it on projects that handle user data or payments.
- github / gitlab - MCP integrations so Claude can read issues, PRs, and repos directly. Worth it if your workflow revolves around GitHub/GitLab.
- Language-specific LSPs (pyright-lsp, typescript-lsp, etc.) - help Claude understand types and definitions. Only install the LSP for the language your project actually uses.
- plugin-dev - the toolkit for building your own plugins. Worth it if you plan to package a capability for your team.
- Superpowers (community) - a bundle of skills for the development workflow (TDD, debug, brainstorm, plan). Worth it if you want to level up the whole workflow, not just one task.
How I pick: start with exactly one plugin that solves your most pressing need (usually commit-commands or one MCP integration), use it for a few days to get used to the namespace and how it is called, and only then add a second. Piling on seven or eight at once usually backfires: you cannot remember which command belongs to which plugin, and your context balloons from the start.
Context-cost note (an optimization tip): every enabled plugin takes up a slice of the session's context/tokens, and the more you install, the more the session "bloats" - Claude has to carry the description of every enabled skill and command. My rule: only enable what you are using for the current project, and turn off the rest. If this topic interests you, see the guide on context and memory management.
Managing, updating, and plugin safety
All management lives in the Installed tab of /plugin: enable/disable, uninstall, and favorite. Right after editing a plugin you are developing, run /reload-plugins to reload it; if the cache is stubborn, add the flag:
/reload-plugins --forceMarketplaces support auto-update - a plugin updates itself when the source ships a new version. That is convenient, but it is also why you should install only from trusted sources: an automatic update brings in new code you have not reviewed. Claude Code also groups plugins you have not touched in a while under "Not used recently," which is handy for periodically clearing out plugins that are quietly eating context you forgot about. Good habit: every few weeks, open the Installed tab and disable what you no longer use.
Safety - do not skip this: a plugin can run arbitrary code with your permissions (hooks, MCP servers, scripts). That means a malicious plugin can do exactly what your account can do. The rule: install only from trusted sources; for plugins outside the official/community marketplaces, read the skill files and MCP config before installing. The official and community marketplaces are vetted; beyond that, you are on your own. For organizations, you can turn on
strictKnownMarketplacesto allow only pre-approved marketplaces.
Free plugins vs a premium kit: when should you pick AgentKit?
Individual (free) plugins are great for one-off needs. But when you want a coherent set of capabilities - many skills, agents, and workflows that fit together - then hand-collecting each plugin, maintaining them, and keeping them from stepping on each other becomes a steady time sink. That is where a pre-curated kit saves you the effort.
To avoid confusion: the "AgentKit" I mean here is the kit for Claude Code (at agentkit.best, used through the ak CLI) - completely different from OpenAI AgentKit (Agent Builder/ChatKit, launched 10/06/2025). And importantly: AgentKit is not in Anthropic's official plugin marketplace - it is a separate kit installed via ak, sitting alongside (not inside) the native plugin system.
On content, AgentKit bundles 108+ skills and 45 agents (17 engineer + 28 marketing), with setup under 5 minutes. Pricing per the site (verify before buying, since it can change): Engineer Kit $99 and Marketing Kit $99 (the site lists no recurring fee), or the Bundle of both for $149; it comes with lifetime updates and a money-back guarantee (the site does not state a specific number of days, so do not trust a figure someone repeats secondhand). You can check AgentKit pricing (20% off via link) to compare for yourself.
Want a coherent set instead of assembling one yourself? If you find you are spending too much time gluing and maintaining individual plugins, a pre-curated kit may be worth a look: the all-in-one kit for Claude Code — now $149 (from $198).
Honest take: if you only need a couple of standalone plugins, the free stuff is enough - no need to spend money. A kit is only worth it when you want a complete set that is curated and maintained for you. To decide with confidence, read the AgentKit review - the kit for Claude Code, and if you want to weigh several options, there is a comparison of the best Claude Code kits.
Frequently asked questions (FAQ)
How is a plugin different from a skill and a subagent?
A plugin is the larger bundle that contains multiple components; a skill and a subagent are two of those components. One plugin can include several skills, several subagents, plus hooks, slash commands, and MCP servers. In short: the plugin is the "box," and skills/subagents are the "items" inside.
Are Claude Code plugins free?
Plugins on Anthropic's official and community marketplaces are free. Any cost comes from a third party - for example, a paid kit like AgentKit, or the service that an MCP server connects to.
Does installing many plugins slow things down or burn tokens?
Yes. Every enabled plugin takes up part of the session's context, so installing many bloats the session and uses more tokens. Only enable plugins you actually use, and lean on the "Not used recently" group to clean up periodically.
Is the official marketplace safe?
The official and community marketplaces are vetted, so they are safer than outside sources. But because a plugin runs code with your permissions, the general rule still holds: install only from trusted sources and read the config before installing, especially with third-party marketplaces.
How do I remove a plugin I do not use?
Open /plugin, go to the Installed tab, select the plugin, then disable it (temporarily off) or uninstall it (remove entirely). Then run /reload-plugins if you need it applied right away in the session.
Is AgentKit a plugin, and how is it different from OpenAI AgentKit?
AgentKit (agentkit.best) is not a plugin in Anthropic's marketplace - it is a separate kit installed through the ak CLI, sitting alongside the native plugin system. It is also completely different from OpenAI AgentKit (Agent Builder/ChatKit) - same name, but two products from two different companies.
Conclusion and next steps
Three takeaways: a plugin is a prepackaged bundle of skills/subagents/hooks/commands/MCP, while a marketplace is the store you browse and install from; you install with /plugin - Discover tab - pick a scope - confirm; and you should only install what you use to save context. To understand the building blocks in depth, read on about what Claude Code skills are; to weigh an all-in-one kit, see the AgentKit review.
Want a more capable Claude Code right now? If you have gone overboard with standalone plugins and want a pre-curated set of skills/agents/workflows that is maintained for you, an all-in-one kit can save real time - just remember to read the terms carefully.