Turn Claude Code into a plugin-powered teammate: a hands-on guide to Anthropic's official plugin directory

Anthropic now maintains an official directory of high-quality Claude Code plugins: 40 internal tools and 14 community plugins you can install with one command. Here is the install flow, the picks worth your first ten minutes, and the trust trade-offs nobody should skip.

Claude Code is already a capable pair programmer, but its real leverage shows up when you stop typing every instruction by hand. Plugins bundle slash commands, subagents, skills, and MCP server configs into installable packages — and for the first year of Claude Code, finding them meant trusting random repos and gists.

That changed with anthropics/claude-plugins-official, Anthropic’s own plugin directory. It sits at 42,296 stars on GitHub, is released under Apache-2.0, and ships 40 internal plugins maintained by Anthropic engineers plus 14 external plugins from partners and the community — each installable with a single command inside Claude Code. This is the hands-on tour: installing, actually using, and knowing when not to install.

What you’ll need

• Claude Code installed and signed in (the directory README points to code.claude.com for the plugin documentation).
• About 10 minutes and a repo or branch you can experiment on.
• A terminal. Every step below runs inside Claude Code itself.

• Repo: anthropics/claude-plugins-official, maintained by Anthropic.
• 42,296 stars on GitHub as of 2026-10-18; 4,507 forks.
• 40 internal plugins (code-review, frontend-design, pr-review-toolkit, hookify, ralph-loop…) plus 14 external plugins (playwright, github, gitlab, linear, telegram…).
• License: Apache-2.0 (check each plugin’s own license file).
• Install syntax: /plugin install {plugin-name}@claude-plugins-official, or browse via /plugin > Discover.
• Trust rule: the README warns — “make sure you trust a plugin before installing” — Anthropic does not control what MCP servers or files a plugin includes.

Step 1 — Open the plugin menu

Type /plugin inside Claude Code. This is the plugin system’s front door: install, browse, and manage plugins from one place. Choose Discover to see the marketplace listing, which is backed by the official directory. You can also jump straight to installation, which is what we’ll do next.

Step 2 — Install your first plugin: code-review

The code-review plugin runs automated pull-request review with multiple specialized agents and confidence-based scoring. Install it:

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

The @claude-plugins-official suffix pins the marketplace, so you know exactly where the code comes from. Once installed, the plugin’s slash command appears:

/code-review

Run it against your current branch and watch it fan out: instead of one generic pass, several reviewers with different briefs (correctness, tests, error handling) score what they find. For heavy PR days, the sibling pr-review-toolkit plugin (command /review-pr) goes deeper — agents specializing in comments, type design, code quality, and simplification.

Step 3 — Add a build skill: frontend-design

Not every plugin is about review. frontend-design is a frontend design skill for UI/UX implementation:

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

Ask Claude to scaffold a page — “build a landing page for a fictional coffee subscription” — and the plugin’s skill steers layout, spacing, and component choices instead of leaving it to defaults. This is the pattern to notice: plugins inject opinion, not just commands.

Step 4 — Pull in an external plugin: Playwright

The external_plugins directory holds partner and community submissions that met the directory’s quality and security bar. The standout is Playwright, Microsoft’s browser-automation and end-to-end testing MCP server:

/plugin install playwright@claude-plugins-official

It hands Claude a real browser: open pages, click elements, fill forms, take screenshots. The loop that justifies the install: generate a page with frontend-design, then have the Playwright-backed agent actually load it, screenshot it, and fix what looks wrong. Other externals worth a look: github, gitlab, linear, context7, and telegram.

Step 5 — Look under the hood before you trust it

Every plugin follows a standard structure — you can read one on GitHub before installing:

plugin-name/
  .claude-plugin/
    plugin.json      # plugin metadata (required)
  .mcp.json            # MCP server configuration (optional)
  commands/            # slash commands (optional)
  agents/             # agent definitions (optional)
  skills/             # skill definitions (optional)
  README.md             # documentation

Two details from the README that matter in practice. First, plugin names are immutable slugs: once published, a name never changes, and renames are handled through a renames map in the marketplace manifest so existing installs auto-migrate instead of breaking with plugin-not-found. Second, if you want to build instead of consume, the directory ships plugin-dev and skill-creator plugins — plus hookify (commands /hookify, /configure, /list) for wiring hooks without hand-editing JSON.

What you built

A Claude Code setup that reviews its own pull requests (/code-review), designs pages with an actual design skill, and drives a real browser through Playwright — all installed from one curated directory in under ten minutes, each addition a versioned package rather than a pasted prompt.

Honest limitations

• Trust is on you. The README is explicit: “make sure you trust a plugin before installing” — Anthropic does not control what MCP servers, files, or other software plugins include, and cannot verify they won’t change. Skim the plugin’s plugin.json, commands, and .mcp.json before installing anything with network access.
• Licenses vary per plugin. The directory itself is Apache-2.0, but the README says to check each plugin’s own license file.
• External plugins are partner-maintained. Quality and security bars exist for inclusion, but updates ship on the maintainer’s schedule, not Anthropic’s.
• Curation, not sandboxing. This directory solves discovery and packaging, not the deeper problem of running third-party agent code safely. Treat plugins like dependencies: pin, review, update.