A plugin in Claude Code is a package that adds skills, sub-agents, hooks, and MCP servers to an agent (Anthropic documentation). The catalog is huge: according to DesignRevision's estimate, there are already over 200 plugins in the official marketplace (DesignRevision review), and thousands in the entire ecosystem.
The main caveat: more plugins do not mean better results. Each of them can execute code with your user's permissions (Claude Code documentation), and also takes up context space. Below are ten plugins worth starting with, and an honest assessment of who doesn't need them.
Contents
📊 Comparison Table
| Plugin | Type | Problem Solved | Verdict |
|---|
| Superpowers | Community | Discipline: plan, specification, tests | Install for large features |
| Context7 | Partner (MCP) | Up-to-date library documentation | Install |
| jdtls-lsp (LSP) | Official | Real code navigation and types | Install |
| code-review | Official | Review of changes before commit | Install |
| security-guidance | Official | Vulnerable patterns in code | Keep enabled |
| GitHub | Partner (MCP) | Repositories, issues, PRs from the session | Install for GitHub |
| Claude Mem | Community | Memory between sessions | For long projects |
| frontend-design | Official | Non-template UI | For frontend |
| Playwright | Partner (MCP) | Browser tests | Situational |
| skill-creator | Official | Creating custom skills | When there are repetitions |
🤖 What are Claude Code Plugins
A plugin is a single directory that you install as a single package. Inside, there can be skills, agents, hooks, and MCP servers (Claude Code documentation). Plugins are distributed through marketplaces: these are catalogs that you add once, and then install plugins from them by name.
The official marketplace claude-plugins-official is available immediately after launching Claude Code. You can view it in the Discover tab of the /plugin command (Anthropic documentation). Everything else is third-party marketplaces: they are added manually.
/plugin install name@claude-plugins-official
⚠️ Important: the plugin marketplace is not Claude Marketplace. These are different things, do not confuse them when searching (Anthropic documentation).
🧭 How Plugins Were Selected
The criteria are simple:
- Origin. Official Anthropic plugins and partner integrations are prioritized. Community plugins are only those with a notable audience and a clear repository.
- Daily usefulness. The plugin should solve a problem that occurs daily: incorrect APIs, loss of context, missed vulnerabilities.
- Cost. How much context and tokens it consumes and what permissions it requires.
The review is based on Anthropic's documentation and open reviews. Descriptions of community plugins are taken from open sources, so check the README before installing. Where I have practical experience, it is highlighted in a separate section.
Left out of the list are Ponytail (a plugin that pushes Claude towards minimal code; mentioned in the DesignRevision review) and Graphify. I haven't found independent tests for them.
🏆 Top 10 Plugins
1. Superpowers
What it is: a community plugin, a set of skills in the form of a development methodology.
What it does: guides Claude through a chain: idea, specification, isolated git worktrees, TDD implementation (DesignRevision review). This means the agent doesn't immediately jump into writing code.
💡 How it might look: you ask to add filters to the API. Claude first clarifies the requirements, writes a specification, and only after your confirmation proceeds to tests and code.
Recommendation: install if you are working on medium and large features.
When it's not needed: for small edits of a few lines, such a ceremony is unnecessary.
What to pay attention to: this is not an Anthropic plugin. Review the repository before installing.
/plugin marketplace add obra/superpowers-marketplace
/plugin install superpowers@superpowers-marketplace
🧪 My experience: a regular agent often starts coding immediately after a request and reports "Done!", even though only part of the scenarios work. Superpowers strictly guides it through stages: brainstorm → specification → plan → tests (TDD) → implementation → verification. When an error occurs, the agent doesn't fix lines randomly: it first collects symptoms, tests hypotheses, and only then makes changes.
Downside: due to mandatory planning and testing, simple edits take more time and tokens than if you had written the code yourself or in a regular dialogue.
2. Context7
What it is: a partner MCP plugin from Upstash.
What it does: fetches library documentation for your version, so Claude doesn't call outdated methods (DesignRevision review).
💡 How it might look: you are working with Spring AI, which changes rapidly. Instead of code for an old API, Claude uses the current documentation and writes for your version.
Recommendation: install if you work with fast-changing frameworks.
When it's not needed: if you rarely go beyond the standard library or work on a frozen internal stack.
What to pay attention to: documentation is fetched from an external service, so do not insert sensitive data into requests.
/plugin marketplace add upstash/context7
/plugin install context7@context7-marketplace
3. LSP Plugin for Your Language (jdtls-lsp for Java)
What it is: an official plugin with a language server. For Java, it's jdtls-lsp (DesignRevision).
What it does: gives Claude go-to-definition, reference search, and type errors instead of text-based guesses (DesignRevision review). There are plugins for TypeScript, Python, Go, Rust, Kotlin, and other languages.
💡 How it might look: you ask to rename a method in a service. Claude finds all calls through the language server, not by text search.
Recommendation: install it обязательно, one plugin for your primary language.
When it's not needed: if there is no language server in the catalog for your language.
What to pay attention to: the server itself must be installed and available in the PATH (DesignRevision).
/plugin install jdtls-lsp@claude-plugins-official
4. code-review
What it is: Official Anthropic plugin.
What it does: Runs parallel agents over a set of changes to find bugs, missed edge cases, and style issues before committing (DesignRevision review).
💡 How it might look: You've finished a feature and run a review before the PR. Agents return a list of remarks before a colleague sees the code.
Recommendation: Install it, especially if you work solo and have no one to review your code.
When it's not needed: If you already have serious automated review in your CI.
What to pay attention to: Multiple agents mean more tokens than a single pass. Check the costs for your repository.
/plugin install code-review@claude-plugins-official
5. security-guidance
What it is: Official Anthropic plugin, enabled by default in recent versions.
What it does: Checks diffs for injections, XSS, SSRF, secret leaks, IDOR, and authorization bypass (DesignRevision review).
💡 How it might look: Claude generated an endpoint where the API key is directly in the code. The plugin flags this before the changes go into the repository.
Recommendation: Keep it enabled.
When it's not needed: There's practically no such case.
What to pay attention to: This is a hint during code generation, not a replacement for a SAST scanner or penetration test.
/plugin install security-guidance@claude-plugins-official
6. GitHub
What it is: Partner MCP plugin in the official marketplace.
What it does: Gives Claude access to repositories, issues, and pull requests directly from the session (DesignRevision review). This is the one Anthropic uses as an installation example in the documentation.
💡 How it might look: You say, "Analyze the issue about a bug in search." Claude reads the description, looks at related PRs, and suggests a fix.
Recommendation: Install it if you live on GitHub. There's a separate plugin for GitLab.
When it's not needed: If your repositories are hosted elsewhere, or if you're content with the standard git in the terminal.
What to pay attention to: Check the token's permissions. Read-only access is sufficient to start.
/plugin install github@claude-plugins-official
7. Claude Mem
What it is: Community plugin for cross-session memory.
What it does: Compresses each session into typed observations and retrieves the necessary ones in the next session (DesignRevision review).
💡 How it might look: A week ago, you decided not to use a certain library. Today, Claude remembers why and doesn't suggest it again.
Recommendation: Install it for long-term projects with a history of decisions.
When it's not needed: If sessions are short and one-off.
What to pay attention to: Memory can accumulate sensitive project details. Check what exactly is stored and where.
npx claude-mem install
8. frontend-design
What it is: Official Anthropic plugin.
What it does: Elevates interfaces from a generic "AI look" to thoughtful typography and hierarchy (DesignRevision review).
💡 How it might look: You ask it to build a landing page. Without the plugin, you get a standard gradient template; with it, you get differently chosen fonts and layout.
Recommendation: Install it if you're building any UI.
When it's not needed: If you only write backend or CLI code.
What to pay attention to: Don't install it with something similar, like ui-ux-pro-max: the functions will be duplicated.
/plugin install frontend-design@claude-plugins-official
9. Playwright
What it is: Partner MCP plugin for browser automation.
What it does: Claude can open pages, click, fill forms, take screenshots, and run E2E tests (DesignRevision review).
💡 How it might look: Claude has built a registration form, opened it in a browser, filled in the fields, and self-verified that the error message appears.
Recommendation: Situational, when you are building or testing a web interface.
When it's not needed: If you don't have a UI.
What to pay attention to: Confirm the exact plugin name for installation in the Discover tab using the /plugin command: the directory changes.
10. skill-creator
What it is: The official Anthropic meta-plugin.
What it does: Asks about your workflow and creates a ready-to-use skill, and can also evaluate and improve existing ones (DesignRevision review).
💡 How it might look: You explain to Claude every time how to format your blog posts. skill-creator turns this explanation into a permanent skill.
Recommendation: Install when you notice repetitive prompts.
When not needed: If you only use plugins created by others.
What to pay attention to: The skill's description determines when Claude will launch it. Check that it doesn't trigger unnecessarily.
/plugin install skill-creator@claude-plugins-official
🔐 Plugin Security
Claude Code documentation divides marketplaces into official and third-party. A plugin from any of them can execute code with your user privileges (Claude Code documentation). Therefore, "official" reduces the risk but does not eliminate it.
What to check before installing a community plugin:
- who the author is and if the repository is active;
- what hooks and MCP servers are included in the plugin;
- where the plugin sends data.
It's better to disable unused plugins. You can do this with /plugin or the command claude plugin disable (Anthropic help).
⚠️ Please note: According to Redwerk's assessment, out of over 9000 plugins, about a hundred are suitable for production (Redwerk). This is an assessment from a secondary source, but the direction is clear: selectivity is more important than quantity.
🎯 Which Set to Choose
| Task | Set | Why |
|---|
| Java Backend | jdtls-lsp, Context7, security-guidance, code-review, GitHub | Precise navigation, up-to-date APIs, security checks and review |
| Frontend and Landing Pages | frontend-design, Playwright, Context7 | Design, browser testing, up-to-date libraries |
| Long Project | Superpowers, Claude Mem | Discipline and memory of decisions |
| Minimum for Start | LSP for your language, security-guidance, code-review | Three plugins without heavy context load |
Start with three to five plugins. Add them one by one and see if the quality improves. If Claude starts behaving strangely, disable the last one added.
About common mistakes when creating MCP servers — in our guide on MCP. About configuring Claude Code — in the CLAUDE.md Best Practices guide
❓ Frequently Asked Questions
How to install a plugin in Claude Code?
Open /plugin and select the Discover tab or use the command /plugin install name@claude-plugins-official. If Claude Code reports that the marketplace is not found, add it via /plugin marketplace add anthropics/claude-plugins-official (Anthropic help).
How many plugins should I install?
Three to five is enough to start. This is my recommendation, not a rule from the documentation: each plugin adds to the context load and risk.
Is it safe to install plugins from third-party marketplaces?
Caution is needed here. Any plugin can execute code with your user privileges, so check the author and repository content (Claude Code documentation).
What is the difference between a plugin, a skill, and an MCP server?
A plugin is a container. It can hold skills, agents, hooks, and MCP servers, which are installed together (Anthropic help).
Are there plugins for Java?
Yes. The official marketplace has jdtls-lsp based on the jdtls server. The server must be installed and available in the PATH (DesignRevision).
📚 Sources
