Claude Code's Plugin Marketplace: Pick Skills Without Wrecking Your Context

Claude Code's plugin marketplace is now a first-class part of the CLI. Starting with v2.1.108 in early May and refined through the 2.1.144 release on May 19, plugins ship through two official directories: claude-plugins-official (Anthropic-curated) and a community directory with reviewed third-party submissions. You install them from inside Claude Code with a slash command — no more cloning repos and editing config files.

What makes this different from a typical extension store is the context cost display. Every plugin in the browse pane shows two numbers: a per-turn token cost (what it adds to every message you send) and a per-invocation cost (what it adds when its skill or command actually fires). Skim those before installing — a plugin that adds 4,000 tokens per turn will eat 40% of a 10k-token budget on day one.

How to Install a Plugin

The install flow is one command. From any Claude Code session:

/plugin install <plugin-name>@claude-plugins-official

# Browse before installing
/plugin browse

# See what you have, with token costs
/plugin list

/plugin browse pulls up an interactive picker with type-to-filter. Each entry shows description, category, and the two context cost numbers. The community marketplace works the same way with @claude-community at the end. Both directories are GitHub-backed, so plugin source is always inspectable before you trust it with your repo.

What a Plugin Actually Bundles

A single plugin can register four things: slash commands, hooks, subagents, and skills. The /radio command Anthropic teased on May 8 is a good example — it adds one slash command that opens a lo-fi music stream and prints the URL when no browser is around. That's a five-line plugin. On the other end, a Terraform plugin might bundle a planning subagent, a hook that runs on file save, three slash commands, and a skill that triggers on .tf files. Same install, very different context footprint.

What's Worth Installing First

From the official directory, three plugins pay for themselves quickly. The github plugin adds gh CLI shortcuts and a PR review subagent. The web-search plugin is a thin wrapper that turns search into a tool call instead of a manual paste. And the testing plugin adds language-aware test runners that detect your framework instead of needing setup. All three sit under 500 tokens per turn.

From the community side, popular picks include language-server bridges for TypeScript, Python, Rust, and Go, plus integrations for Linear, Asana, and Firebase. The Linear plugin in particular is one of the few that genuinely justifies its 2,000-token per-turn cost if you actually live in Linear all day.

How to Uninstall Cleanly

Plugins are scoped to either the current project or your global config. /plugin remove takes a flag to control which:

# Remove from current project only
/plugin remove <plugin-name>

# Remove from global install (affects all projects)
/plugin remove <plugin-name> --global

# See what's loaded for this project vs globally
/plugin list --scope all

If a plugin breaks your session — bad hook, runaway subagent, MCP server that won't shut down — start Claude Code with --no-plugins to bypass everything. From there you can /plugin remove the offender. This is the cleanest recovery path when something installed from the community directory misbehaves.

The Real Reason This Matters

Before the marketplace, sharing a Claude Code workflow meant DM'ing someone a CLAUDE.md snippet and hoping they could replicate it. Now there's a one-line install that brings the same commands, hooks, and skills to anyone's setup. Workflows are becoming portable. The context cost display makes it possible to ship those workflows without quietly destroying the recipient's token budget — which is what made browser extension stores feel safe a decade ago.