Roundup · Updated July 2026

The 9 Best Claude Code Plugins in 2026

Claude Code's plugin ecosystem went from a handful of experiments to thousands of entries in about a year, which means the problem is no longer finding plugins — it's finding the ones worth the context they consume. This is the short list: plugins we actually keep installed, organized by the job they do, with an honest note on the one we built ourselves.

TL;DR — the picks

How we evaluate

1. Does it change what ships? A plugin has to improve the code or the loop, not just add slash commands you forget exist.
2. Is it worth its context? Every plugin loads instructions into your session. Bloated ones make the agent worse at everything else.
3. Transparent bias: Vibejar is our product. It's marked below — judge for yourself.

What is a Claude Code plugin?

A Claude Code plugin is a shareable package that extends the agent with any mix of five things: skills (instructions invoked as /plugin-name:skill-name), subagents, hooks that fire on events, MCP servers that add tools, and LSP servers for code intelligence. Skills are the part people mean most often — a skill is a Markdown playbook the agent loads for a specific kind of task. A plugin is the distribution wrapper that lets a team or the community install that playbook in one command.

How installation works

Plugins install from marketplaces — git repos with a plugin catalog. Anthropic's official marketplace is registered automatically the first time you start Claude Code, so browsing is one command:

/plugin

That opens the manager. To add the reviewed community marketplace and install from it:

/plugin marketplace add anthropics/claude-plugins-community
/plugin install some-plugin@claude-community

Any GitHub repo with a marketplace manifest works the same way via /plugin marketplace add owner/repo, and you can test a local plugin with the --plugin-dir flag before publishing anything. After installing, run /reload-plugins or restart the session.

The list

1. Superpowers

The most popular community plugin, and it earns it. Superpowers imposes a workflow on the agent — brainstorm, plan, test, then build — which is exactly the discipline vibe-coded projects lack by default. It's the closest thing to a senior engineer's process in a box.

• Plan-and-spec-first workflow skills
• Pushes the agent to verify work instead of declaring it done
• Free and open source

Best for: anyone whose agent keeps sprinting past the thinking step.

2. Context7

Agents hallucinate APIs from their training data; Context7 fixes that by pulling current, version-specific documentation for the library you're actually using into context. The difference shows up immediately in framework-heavy work — fewer invented methods, fewer deprecated patterns.

• Fresh docs on demand via MCP
• Version-aware, so Next.js 16 answers aren't Next.js 13 answers
• Free tier covers solo use

Best for: projects on fast-moving frameworks.

3. LSP plugins (official)

Anthropic ships language server plugins for TypeScript, Python, Rust, Go, and other major languages. They give the agent what your editor has: go-to-definition, references, and live diagnostics instead of grep and guessing. Boring, foundational, install-and-forget.

• Real code intelligence for navigation and refactors
• Type errors surface as the agent works, not at build time
• One install per language

Best for: everyone. This should be the default.

4. frontend-design (official)

Anthropic's own plugin for UI work, aimed squarely at the "AI-generated look" problem — the generic gradients-and-cards output every agent produces unprompted. It steers toward restrained, production-grade interfaces and is the cheapest design upgrade an agent project can get.

• Design guidance the agent applies while building UI
• Reduces the default-template sameness
• Free, in the official marketplace

Best for: solo builders shipping UI without a designer.

5. pr-review-toolkit (official)

A set of review agents that read a pull request the way a careful colleague would — correctness, tests, and comment quality — instead of the rubber stamp agents default to. Useful even solo: run it on your own PRs before merging what the agent wrote.

• Structured, multi-pass PR review
• Catches silent rewrites before they merge
• Free, in the official marketplace

Best for: repos where the agent writes most of the diffs.

6. commit-commands (official)

Small and frictionless: slash commands for committing, branching, and pushing with sane messages. It mostly removes the "agent writes a paragraph-long commit message" problem and keeps history readable when you're shipping dozens of agent commits a day.

• One-command commit and push flows
• Consistent, scannable commit messages
• Free, in the official marketplace

Best for: high-volume agent workflows.

7. claude-mem

Claude Code forgets everything between sessions; claude-mem doesn't. It persists decisions, context, and project knowledge across sessions, which matters most on long-running solo projects where you are the only institutional memory.

• Persistent memory across sessions
• Automatic capture, searchable later
• Free and open source

Best for: projects you return to weekly rather than daily.

8. security-review (official)

Anthropic's security review plugin audits code for the vulnerability classes agents introduce quietly: injection, auth gaps, exposed secrets. Vibe-coded apps fail security checks nobody ran — this is the one that runs them. Point it at your diff before shipping anything that touches user data.

• Security-focused review pass with severity ratings
• Built and maintained by Anthropic
• Free

Best for: anything with auth, payments, or user data.

9. Vibejar — our product

The plugins above make Claude Code better at writing code. Vibejar feeds it the bugs. You screenshot a bug in your own iOS or Android app, circle what's wrong, and it lands in a queue. The agent-side install — one command — gives Claude Code a skill and CLI to list captures, claim one, read the annotated screenshot, ship the fix, and report back with a PR and before/after proof. It also works with Codex and Cursor, so the loop isn't married to one agent. Full disclosure: we built it — it's our answer to the vibe debugging problem, and the workflow is described in the screenshot-to-fix loop.

• Annotated mobile bug captures delivered straight to the agent
• Claim-one-at-a-time queue, safe across parallel sessions
• Fixes end as PRs with before/after proof

Pricing: $88 one-time.

Best for: vibe coders with a mobile app — the screen Claude Code can't see on its own.

Frequently asked questions