9 Best Claude Code Plugins in 2026 (Plus a Bonus Tip)
TL;DR
A Claude Code plugin is a self-contained bundle of skills, subagents, hooks, MCP servers, and slash commands installed from a marketplace with one command — /plugin install <name>@<marketplace> — rather than something you assemble by hand in .claude/.
The public plugin catalog at claude.com/plugins lists 200+ plugins, spanning Anthropic's curated official marketplace, an automated-review community marketplace, and installable third-party listings, with install counts and an "Anthropic verified" badge shown for each.
The most-installed plugins right now skew toward frontend code generation, agent/skill development, and AI-assisted code review — not toward niche language servers or single-purpose utilities, which says more about how teams actually use Claude Code day to day than any single feature list does.
The official marketplace's security-guidance, commit-commands, and pr-review-toolkit plugins cover the workflow most teams reach for first: catching risky code changes, generating commit messages and PRs, and running specialized review agents against open pull requests.
No official or community plugin ships a general-purpose, live web-data-access MCP server out of the box — teams that need Claude Code to fetch, render, and clean external pages on demand still have to wire up an MCP server themselves, typically against a crawling API.
Installing a plugin can add meaningful token overhead every turn; Claude Code shows a "Context cost" estimate before you install specifically so you can weigh that against what the plugin actually does.
Plugins run with your full user privileges, and Anthropic explicitly does not vet what an MCP server, hook, or bundled script actually does — the verified badge only reflects the curator's own quality process, not proof of safety.
At a Glance: 9 Claude Code Plugins Worth Installing (Plus a Bonus Tip)
Rank
Plugin
Category
Source
Best for
1
Frontend DesignCraft
Code generation
Anthropic (verified)
Frontend UI that doesn't look like default AI output
2
Superpowers
Agent development
Community
Brainstorming, subagent design, TDD, and skill authoring in one bundle
3
Code Review
Code quality
Anthropic (verified)
Confidence-filtered AI review on pull requests
4
Context7
Documentation / MCP
Upstash
Pulling version-specific library docs directly into context
5
Skill Creator
Agent development
Anthropic (verified)
Building, evaluating, and benchmarking your own skills
6
Code Simplifier
Code quality
Anthropic (verified)
Cleaning up recently changed code without altering behavior
7
Commit Commands + PR Review Toolkit
Git workflow
Anthropic (verified)
Commit messages, PR creation, and specialized PR review agents
8
Security Guidance
Security
Anthropic (verified)
Flagging injection, XSS, and unsafe patterns as Claude writes code
9
Playwright
Testing
Microsoft
Browser automation and end-to-end test generation
Bonus
Custom web-data-access plugin (Nstproxy Crawl)
Data access / MCP
Self-built
Live web fetching, RAG ingestion, and agent research the catalog doesn't cover
What Actually Counts as a Claude Code Plugin
A Claude Code plugin is a directory with a .claude-plugin/plugin.json manifest that can bundle skills, subagents, hooks, MCP server configurations, and LSP server configurations into one installable unit, according to Anthropic's own plugin documentation. That's a narrower definition than "anything that extends Claude Code" — a single MCP server entry you paste into .mcp.json yourself isn't a plugin until someone packages it with a manifest and lists it in a marketplace. A marketplace is just a catalog: adding one (/plugin marketplace add owner/repo) registers it with Claude Code, and installing a specific plugin from it (/plugin install name@marketplace) is a separate second step. Plugin skills are namespaced by the plugin name, so a plugin called commit-commands exposes its skill as /commit-commands:commit, which is how Claude Code avoids collisions when two plugins ship a skill with the same short name.
How We Evaluated These Plugins
Ranking "best" plugins honestly means separating three different signals that generic roundups tend to blur together: whether a plugin comes from Anthropic's own curated official marketplace (the claude-plugins-official repository) versus the automated-review community marketplace, how many real installs it has on the public claude.com/plugins catalog, and whether it fills a category the catalog doesn't cover at all. Anthropic verified status reflects Anthropic's own curation process for the official marketplace — it is not a security audit, and Anthropic's plugin documentation says directly that it "doesn't control what MCP servers, files, or other software are included in plugins and can't verify that they work as intended." Install counts, by contrast, are a real usage signal pulled straight from the public catalog page. The 9 ranked entries below are ordered primarily by installs within their category, with verified status as a tiebreaker. After the ranked list, a bonus tip covers a different kind of gap: pulling live external web content into a Claude Code session on demand is a job no catalog listing currently does off the shelf, so it doesn't fit a ranking built on install counts — but it's common enough that it's worth covering anyway.
Take a Quick Look
Install counts can't rank a capability the catalog doesn't offer yet — if your team keeps hitting that wall for live web data, skip to the bonus tip below for how Nstproxy Crawl fills it.
1. Frontend DesignCraft: Best for Frontend UI That Doesn't Look Default
Frontend DesignCraft is Anthropic's own verified plugin for generating frontend code, and it's the single most-installed plugin in the public catalog at over 1.13 million installs. Its stated purpose is producing "production-grade frontends with distinctive design" that "avoids generic AI aesthetics" — a direct response to the recognizable, sameish look that comes out of a lot of default AI-generated UI code. It's the right pick for teams whose Claude Code usage skews toward building or iterating on user-facing screens, where visual distinctiveness matters as much as functional correctness.
2. Superpowers: Best for Agent Development in One Bundle
Superpowers is a community-marketplace plugin with over 1 million installs that bundles brainstorming, subagent development with built-in code review, debugging workflows, test-driven development, and skill authoring into a single install. Its install count second only to Frontend DesignCraft is notable precisely because it isn't an Anthropic-verified official plugin — it's a case where real usage outpaced official curation, which is exactly the kind of signal a pure "official marketplace only" roundup would miss entirely. It suits teams that want one plugin covering the full arc of extending Claude Code itself, rather than installing five narrower plugins for the same set of tasks.
3. Code Review: Best for Confidence-Filtered PR Review
Code Review is Anthropic's verified plugin for AI-assisted pull request review, built around specialized review agents and what the catalog describes as "confidence-based filtering" — surfacing the issues a reviewer is more likely to actually act on rather than flooding a PR with every possible nitpick. With over 438,000 installs, it sits well behind the top two general-purpose plugins but ahead of every other single-purpose code-quality tool in the catalog, which tracks with how central PR review is to most teams' actual Claude Code usage.
4. Context7: Best for Version-Specific Library Documentation
Context7 is Upstash's MCP-server plugin for pulling live, version-specific documentation and code examples directly from source repositories into Claude's context, with over 417,000 installs. It solves a specific, recurring problem: a model's training data goes stale the moment a library ships a breaking change, and Context7 closes that gap by fetching current docs at request time instead of relying on what Claude already "knows" about a package version. It's a strong pick for teams working against fast-moving libraries or frameworks where outdated API guidance costs real debugging time.
5. Skill Creator: Best for Building and Benchmarking Your Own Skills
Skill Creator is Anthropic's verified plugin for creating, updating, evaluating, and benchmarking custom skills, with over 385,000 installs. It's a meta-tool in the most literal sense — a plugin for building the kind of custom .claude/skills/ extensions (or full plugins) that this entire list is made of — and it's the natural next step for a team that's outgrown the catalog's off-the-shelf options and wants to formalize its own internal skill library with actual evaluation data behind it, rather than ad hoc prompt files nobody remembers to update.
6. Code Simplifier: Best for Cleaning Up Recently Changed Code
Code Simplifier is Anthropic's verified "code clarity agent," with over 346,000 installs, that simplifies and refines recently modified code while explicitly preserving functionality and consistency with the surrounding codebase. It's scoped narrower than Code Review — it's about readability and structure in code that was just touched, not catching bugs or security issues — which makes it a natural pairing with Code Review rather than a substitute for it.
7. Commit Commands + PR Review Toolkit: Best for End-to-End Git Workflows
Commit Commands and PR Review Toolkit are two separate Anthropic-verified official plugins that cover opposite ends of the same workflow, and teams tend to install them together. Commit Commands adds git workflow skills for staging, generating commit messages, pushing, and opening pull requests — invoked as /commit-commands:commit once installed. PR Review Toolkit adds specialized agents that review comments, tests, error handling, type correctness, code quality, and simplification opportunities on an already-open PR. Installed together, they cover a change from "commit this" through "is this PR actually ready," which is why they're grouped here rather than ranked as two competing single-purpose tools.
8. Security Guidance: Best for Catching Risky Patterns as Claude Writes Code
Security Guidance is Anthropic's verified official plugin that reviews each change Claude makes for common vulnerability classes — command injection, cross-site scripting, and other unsafe patterns — and instructs Claude to fix what it finds within the same session, rather than waiting for a separate security review pass. It's a lower-installed pick than the code-quality plugins above it, but it's the one entry on this list that actively catches a category of mistake most teams would otherwise only find in a later audit or, worse, in production.
9. Playwright: Best for Browser Automation and End-to-End Testing
Playwright is Microsoft's official browser-automation plugin, giving Claude Code the ability to drive a real browser for end-to-end test generation and interaction testing. It's the one entry on this list that isn't from Anthropic's own marketplace or a documentation/data-access category — it's here because browser-driven testing is a distinct, recurring need that none of the code-quality or git-workflow plugins above it address, and because Playwright's install base reflects genuine adoption rather than default-marketplace placement.
Bonus Tip: Build a Web-Data-Access Plugin with Nstproxy Crawl
None of the 9 ranked plugins above wrap a general-purpose page-fetching-and-rendering API the way a dedicated MCP server would, which is worth flagging separately since it doesn't fit a ranking built on install counts — no catalog listing does this job today, so there's no install number to rank it by. Claude Code's own MCP documentation describes connecting to "hundreds of external tools and data sources," and a crawling API is squarely the kind of external tool that pattern is built for. Nstproxy Crawl is a good fit for teams that keep hitting this gap: it's a REST API that turns a URL into clean Markdown, cleaned HTML, links, or a screenshot, with JavaScript rendering and Nstproxy's own proxy-backed access bundled in, and official Node.js, Python, and Go SDKs so wiring it into an MCP server doesn't mean writing an HTTP client from scratch. It fits teams building AI agents that need to read the live web, RAG pipelines that ingest external pages on a schedule, or competitive and pricing research that a static documentation-lookup plugin can't touch — and it's the wrong tool if you just need occasional web search, since Claude Code's built-in web search already covers that without any extra setup.
Site-level and single-page crawling — POST /api/v1/crawl discovers and scrapes an entire site within the maxDepth and maxPages bounds you set, while POST /api/v1/crawl/scrape handles one URL synchronously or asynchronously.
Multiple structured output formats — Markdown, cleaned HTML, links, and screenshots come back from the same request, so an MCP server built on it can hand Claude whichever format a given task needs.
Pay-as-you-go pricing with no subscription requirement — plans currently start at $79/month for the Starter tier at $1.00 per 1,000 URLs, alongside a free tier billed per request, so a small internal MCP server doesn't require committing to enterprise pricing.
Building this as an MCP plugin takes more setup than clicking install on a catalog entry — you're writing (or adapting) an MCP server that calls the Crawl API and exposing it through your own .mcp.json or a plugin manifest, not installing something pre-packaged. That tradeoff is the point: it buys you a capability the current catalog doesn't offer prepackaged, at the cost of the five-minute setup a listed plugin gives you.
Take a Quick Look
If your Claude Code workflow needs to pull live pages, not just search snippets, from sites outside the marketplace, Nstproxy Crawl gives you the fetch-and-render API to build that MCP server yourself.
Side-by-Side: Category, Source, and Setup Snapshot
Plugin
Category
Verified
Setup effort
Frontend DesignCraft
Code generation
Yes
One-command install
Superpowers
Agent development
No
One-command install
Code Review
Code quality
Yes
One-command install
Context7
Documentation / MCP
No
One-command install, API key optional
Skill Creator
Agent development
Yes
One-command install
Code Simplifier
Code quality
Yes
One-command install
Commit Commands + PR Review Toolkit
Git workflow
Yes
One-command install (two plugins)
Security Guidance
Security
Yes
One-command install
Playwright
Testing
N/A (Microsoft)
One-command install, browser binaries required
How to Choose the Right Plugins for Your Workflow
Start from the gap in your current setup rather than installing everything on this list at once — each plugin the plugin manager shows you carries a "Context cost" estimate for a reason, and stacking unused plugins adds token overhead to every single turn without adding capability you're using. Teams shipping a lot of UI work get the most immediate value from Frontend DesignCraft; teams doing heavy git/PR churn should pair Commit Commands with PR Review Toolkit rather than picking one; and teams building their own internal tooling on top of Claude Code get more long-term leverage from Superpowers and Skill Creator than from any single-purpose utility. If your team's Claude Code usage already involves pulling external web content — competitor pages, pricing pages, documentation that isn't in Context7's library index — that's the signal to build the custom MCP plugin described in the bonus tip above rather than trying to force a documentation-lookup or search-only tool into that job.
Common Use Cases by Team Type
Frontend and product teams lean on Frontend DesignCraft plus Code Review to keep UI work both distinctive and reviewed. Platform and infrastructure teams tend toward Security Guidance, Commit Commands, and PR Review Toolkit as their standard baseline, since those three plugins map directly onto a typical change-review pipeline. Teams building AI agents or internal tooling on Claude Code itself get the most out of Superpowers and Skill Creator, since both are meta-tools for extending Claude Code rather than tools for a specific language or framework. Teams doing research, monitoring, or RAG-style data collection are the ones who consistently hit the gap the bonus tip describes, since Context7 covers library documentation specifically and doesn't fetch arbitrary external pages.
Conclusion
The plugins with the highest install counts in the public catalog right now cluster around frontend code generation, agent and skill development, and AI-assisted review — not around the long tail of single-language utilities a "complete list" might suggest matters equally. Anthropic's own official marketplace covers git workflows, security review, and code quality solidly, Upstash and Microsoft each fill a specific external-integration gap, and the one category none of them address off the shelf is general-purpose live web access, which is where a custom MCP plugin built on an API like Nstproxy Crawl fits. Pick plugins based on where your team's actual Claude Code sessions spend their time, not based on installing the longest possible list.
Not automatically — Anthropic's own plugin documentation states plainly that plugins "can execute arbitrary code on your machine with your user privileges" and that Anthropic "doesn't control what MCP servers, files, or other software are included in plugins and can't verify that they work as intended," so an "Anthropic verified" badge reflects the official marketplace's curation process, not a security audit of every plugin's code.
Q: Do plugins cost extra, either in dollars or context tokens?
Most catalog plugins are free to install, but they aren't free in context: Claude Code shows a "Context cost" estimate before you install a plugin because its skills, agents, and MCP tool definitions add tokens to every turn whether or not you use them that session, so installing plugins you rarely touch has an ongoing cost even at zero dollars.
Q: How do I install a plugin from the official marketplace?
Run /plugin marketplace add anthropics/claude-plugins-official if the marketplace isn't already registered (Claude Code adds it automatically on first interactive launch), then install a specific plugin with /plugin install <plugin-name>@claude-plugins-official, for example /plugin install security-guidance@claude-plugins-official.
Q: Can I build and share my own Claude Code plugin?
Yes — a plugin is just a directory with a .claude-plugin/plugin.json manifest plus any combination of skills/, agents/, hooks/, .mcp.json, or .lsp.json, which you can test locally with claude --plugin-dir ./my-plugin and then distribute through your own marketplace or submit to Anthropic's community marketplace for review.
Q: Is there a way to give Claude Code live web access as a plugin?
Not as a single official catalog listing today — the closest official building block is Claude Code's built-in web search, which doesn't fetch and render arbitrary external pages the way a dedicated crawling API does, so teams that need that capability build a custom MCP server against an API like Nstproxy Crawl and package it as their own plugin.
Q: What's the difference between a plugin and just adding an MCP server directly?
A raw MCP server entry in .mcp.json only works in the project where you added it and has to be re-added by hand anywhere else you need it; wrapping the same server in a plugin with a manifest lets you install it with one command, share it through a marketplace, and get it namespaced automatically so it doesn't collide with another plugin's tools of the same name.
Ivy Lin
Aug. 12th 2026
110M+ real IPs with 99.9% access success
Blazing-fast average response ~0.5s for high-concurrency tasks
From only $0.1/GB
Get immediate access to premium residential, datacenter, IPv6 and ISP proxy pools.
