Best Claude Plugins
Plugins and integrations that extend Claude's capabilities.
What are Claude Code Plugins?
A Claude Code plugin is a distributable bundle that adds a coordinated capability, typically a combination of skills, subagents, MCP servers, hooks and commands, installed and updated as a single package. Plugins make it possible to ship a complete workflow (security audit toolkit, design-system enforcer, monorepo orchestrator…) without users having to wire each piece by hand.
Plugins are the highest-level primitive in the Claude Code ecosystem. Anthropic's plugin marketplace lets developers publish, version and update plugins; users install with claude plugin install <plugin>. Under the hood the plugin can register MCP servers, drop subagent files, configure hooks and inject context, all from one declarative manifest.
Below: every open-source plugin we've indexed, with star counts, last activity and Trust Score. Filter by use case to narrow down (security, dev-tools, content, marketing…).
Questions about Plugins
How do I install a Claude Code plugin?+
Run claude plugin install <plugin-name> if it's in the official marketplace, or clone the repo and run the installer (most ship a npm-style install script). The plugin's README on its repo page lists exact steps; on ClaudeWave the install block is auto-generated.
Plugin vs skill vs MCP server, when do I pick which?+
Use a skill for a single workflow you reuse. Use an MCP server when Claude needs to call an external system. Use a plugin when you want to ship multiple of the above as one cohesive package, e.g. a 'security suite' plugin that bundles 3 subagents, 2 MCP servers and a hook.
Can plugins update automatically?+
Yes if installed from the marketplace. Run claude plugin update to refresh. Manually-installed plugins follow whatever update mechanism the maintainer ships (git pull, npm update, etc.).
How do I publish my own plugin?+
Create a manifest (plugin.json) declaring the skills, subagents, MCP servers and hooks the plugin contributes. Test locally with claude plugin install ./. Push to GitHub. Submit to the Anthropic marketplace if you want first-party discovery.
Are plugins sandboxed?+
Partially. Plugins inherit the same permission model as standalone skills/agents/MCP servers, Claude still asks for tool confirmation. There's no full container isolation. Treat plugins like any third-party install: read the manifest, check the Trust Score, prefer maintained projects.