The Best Claude Code Plugins in 2026 (and How to Install Them in One Command)
A plugin is the packaging format Claude Code was missing: one install that carries skills, subagents, slash commands, hooks, and MCP servers together, versioned and updatable, instead of fifteen folders you copied from GitHub in March and never touched again. The catch is that plugins are also the easiest thing in the ecosystem to publish badly — a README, four thin prompts, and a marketplace listing. Here is what's actually worth installing, ranked, plus how to judge anything not on the list.
Our #1: the AgentsKit kits
Disclosure first: we build this. Read the rest with that in mind — and then judge it on the criteria in the section below, which is how we'd want you to judge ours.
The case for it is coverage plus maintenance. Most plugins solve one slice: a review agent, a test writer, a docs helper. The problem with a slice is that your setup ends up as eight plugins from eight authors, with overlapping triggers, contradictory conventions, and three abandoned repos. AgentsKit ships 89 agents, 103 skills, and 181 commands as one coherent library across engineering and marketing — descriptions tuned so things actually fire, deterministic work in bundled scripts, and one place to update when Claude Code changes.
If you want the honest disqualifier: it's paid, and if your needs are narrow — you want a review skill and nothing else — a free single-purpose plugin is the better buy. The kits earn their place when you're running Claude Code daily across a real codebase or a real content operation. See the full contents →
2. Anthropic's official plugins and document skills
The first-party ones set the standard, particularly for documents: PDF, Word, and spreadsheet handling that produces real files instead of markdown pretending to be a deliverable. They're free, well-built, and demonstrate the pattern worth copying — the model decides, bundled scripts execute the mechanical parts. Install these before anything from a stranger.
3. Language and framework plugins
The good ones encode idiom, not syntax. A strong Next.js or Rails or Django plugin knows your framework's file conventions, its testing idiom, and the six things people get wrong in it — so Claude stops writing generically correct code that looks foreign in your repo. Install the one matching your primary stack; skip the rest, since inactive plugins still occupy description space.
4. Review and security bundles
Severity-ranked review, injection and authz checks, secret scanning, dependency audits. This is the category where a consistent checklist beats a smart improvisation every time, which is exactly what a plugin is good at. It also pairs with hooks: the review runs, and a hook makes sure it ran.
5. Marketing and content plugins
Underrated, because most people still think of Claude Code as a developer tool. SEO writing, brand-voice enforcement, campaign planning, and performance reporting work unusually well here for the same reason coding does: the work is files in a folder, and an agent that can read and write the whole folder beats a chat window you paste into. More on that in Claude Code for marketers.
How to judge a plugin in sixty seconds
| Check | Good sign | Walk away |
|---|---|---|
| Descriptions | Name concrete triggers and file types | "You are an expert in…" and nothing else |
| Content | Checklists, steps, worked examples, scripts | Adjectives and encouragement |
| Maintenance | Commits since the last Claude Code release | Last touched eight months ago |
| Scope | Says what it does not do | Claims to do everything |
| Safety | Readable source, no surprise network calls | Obfuscated scripts, broad hook permissions |
That last row matters more for plugins than for standalone skills. A plugin can register hooks and MCP servers, which means installing one can grant it the ability to run commands around your tool calls. Read what you install, the same way you'd read a dependency that ships a postinstall script.
Installing and managing them
Plugins come from marketplaces, which are just git repos with a manifest. You add a marketplace, then install from it — /plugin marketplace add owner/repo followed by /plugin install name@marketplace inside Claude Code. Project-level plugin config lives in your repo, so a teammate cloning it gets the same setup rather than a message from you explaining the setup.
Then actually verify. Ask Claude which skills and agents it has available, phrase a request that should trigger one naturally, and watch whether it fires. Half of "this plugin doesn't work" is a description problem, not an install problem — our free SKILL.md validator and the triggering guide cover the fix.
What not to install
- Prompt-pack plugins. Fifty prompts in a trench coat. Prompts belong in a library you copy from, not in your trigger space.
- Anything duplicating built-ins. File reading, git, running commands — Claude Code already does these natively.
- Overlapping bundles. Two plugins with review agents means two agents competing for the same trigger and neither winning cleanly.
One install instead of eight: AgentsKit bundles 89 agents, 103 skills, and 181 commands across engineering and marketing — triggers tuned, scripts included, maintained as Claude Code changes. See the full contents →