Anthropic opened a Claude Code plugin store, and the first thing on the page is a warning
The short version
Claude Code now has a first-party plugin directory: anthropics/claude-plugins-official. You install a plugin with one command and it can carry slash commands, agents, skills and MCP servers. Anthropic states plainly that it does not control or verify what is inside a plugin, so installing one is trusting third-party code with your machine.
What changed
Anthropic published a curated directory of Claude Code plugins at anthropics/claude-plugins-official, current as of 17 October 2026. It replaces the earlier situation where plugins lived in scattered community repositories and you found them through a Discord message or a blog post.
The repository has two halves: /plugins for the ones Anthropic builds and maintains, and /external_plugins for third-party submissions from partners and the community.
How to install one
Two routes, both from inside Claude Code:
/plugin install {plugin-name}@claude-plugins-official installs a named plugin directly. /plugin then Discover opens a browsable list instead, which is the better choice if you do not already know what you want.
External plugins go through a submission form and are supposed to meet quality and security standards before they appear. Anthropic does not publish what those checks are, which matters for the next section.
What a plugin actually carries
| Component | What it adds |
|---|---|
plugin.json | Metadata. Required for every plugin |
.mcp.json | MCP server configuration, meaning new network connections |
commands/ | Slash commands you type |
agents/ | Agent definitions that run on their own |
skills/ | Skill files that load into context |
That second row is the one to read twice. A plugin can point Claude Code at an MCP server it does not own, which means a new connection from your machine to someone else’s service, with your prompts and whatever files the agent reads.
The catch, in Anthropic’s own words
The directory README carries a warning above everything else: make sure you trust a plugin before installing it, because Anthropic does not control what MCP servers, files or other software are included in plugins, cannot verify that they work as intended, and cannot verify that they will not change later.
That is an unusually direct disclaimer for a first-party directory, and it should set your expectations. Being listed is not an audit. It is a curation decision by a company that has said it is not auditing.
The second catch is quieter. Plugin names are immutable slugs: once published, a name cannot change, because renaming it breaks every existing install with a plugin-not-found error. Renames are handled through a renames map that rewrites the old slug on your next sync. The effect is stability for you, and a directory that accumulates dead weight rather than cleaning up.
Plugins versus skills, which people keep asking
A skill is a file that teaches the agent how to do something. A plugin is a package that can contain skills plus commands, agents and servers. If all you want is a reusable instruction set, a skill folder is the smaller thing to maintain.
There is also the cost angle. Plugin and skill descriptions load into context, and the top autocomplete question about plugins right now is which ones reduce token usage rather than increase it. Installing six plugins to save time can end up costing more per session than doing the work once.
If you are going to install one
Start with /plugins and pick something from Anthropic’s own list before you touch external submissions. Read the plugin’s repository, not just its marketplace description. Check whether it declares an MCP server, and if it does, read which domain that server talks to.
Claude Code has no free tier, so every plugin you add is running on a paid subscription or API credits. Test one plugin for a week before adding a second.
Questions people ask
How do I install Claude Code plugins?
Run /plugin install name@claude-plugins-official inside Claude Code, or open /plugin and choose Discover to browse the directory.
Are official Claude Code plugins safe?
They are curated, not audited. Anthropic states it does not control or verify the MCP servers, files or software a plugin contains.
What is the difference between a plugin and a skill?
A plugin is a package that can include skills, slash commands, agents and MCP server configuration. A skill on its own is just an instruction file.
Sources
github.com/anthropics/claude-plugins-official (README, read 20 October 2026); Claude Code plugin documentation at code.claude.com/docs; Claude Code release notes.
Read from the primary source on 20 October 2026. Prices and terms change, so check before you commit.