Claude Code Plugins: Install, Scope and Remove (2026)

4 min read

How to install a Claude Code plugin with /plugin install or claude plugin install, pick a scope (user, project or local), and check its context cost. Commands from the official plugin docs, read 2026-11-14. Includes a memory section.

What is a Claude Code plugin?

A Claude Code plugin is a folder that bundles skills, subagents, hooks and MCP servers, so you can install them as one unit. Most plugins come from a marketplace, a catalog that lists plugins and where to get each one. The folder has a manifest at .claude-plugin/plugin.json that gives it a name and version.

Use a plugin when you want several of these pieces packaged together, with updates from the marketplace. A single skill or MCP server you can also set up on its own. For those, see Claude Code MCP and Claude Code skills.

How do you install a plugin?

Start Claude Code in your project, then run one of these:

Where you run itCommandWhat happens
Claude Code session/plugin install <name>@<marketplace>Opens the plugin’s details. You review it and pick a scope before anything installs.
Claude Code session/pluginOpens the panel on the Discover tab, where you can search plugins from your marketplaces.
Shellclaude plugin install <name>@<marketplace>Installs from the command line.

The official marketplace is claude-plugins-official. Claude Code adds it the first time you start an interactive terminal session. It does not add community marketplaces on its own. For any other marketplace, add it first with claude plugin marketplace add, then install.

1claude plugin install commit-commands@claude-plugins-official
2claude plugin list

After install, type / and look for the plugin’s skills under its name, in the form /<plugin>:<skill>. If a plugin’s server needs setup, or the plugin shows a load error, the Errors tab in /plugin gives the reason. After you change a plugin’s files, run /reload-plugins.

Which install scope should you pick?

The scope decides who the plugin is enabled for.

ScopeEnabled forUse it for
UserYou, in every project on this machineTools you use everywhere
ProjectEveryone who works in this repository (set in the committed .claude/settings.json)A setup the team should share
LocalYou, in this repository onlyTrying a plugin on one project

Claude Code reads the same settings files in the terminal, the desktop app’s local sessions and the VS Code extension. A cloud session, including one in the browser at claude.ai/code, does not load plugins from your local settings.

How do you check, turn off or remove a plugin?

1claude plugin list                          # name, version, scope and status
2claude plugin disable <name>@<marketplace>  # turn off, keep installed
3claude plugin enable <name>@<marketplace>   # turn back on
4claude plugin update <name>@<marketplace>   # pull the latest version
5claude plugin uninstall <name>@<marketplace>

Disabling is the quick way to stop a plugin’s cost without losing its setup. The Installed tab in /plugin has a “Not used recently” group that lists plugins you could turn off.

What does an enabled plugin cost?

An enabled plugin is part of every session, not only the ones where you use it. Two costs matter:

  • Context on every turn. For each skill, agent and command that Claude can invoke on its own, the name and description sit in context all the time. Those tokens leave less room for your work.
  • Processes and hooks. The plugin’s MCP servers run next to each session, and its hooks fire on their events.

Plugins in Anthropic’s official marketplace show two estimates in /plugin before you install: the cost on every turn, and the cost once Claude loads a skill or agent. Plugins from a local or custom marketplace do not show them.

A plugin runs code with your user rights. Read the plugin’s components before you install it, and check the security notes in the docs.

Memory: does a plugin give Claude Code a memory?

A plugin gives Claude memory only if it ships a part that saves facts and returns them later. Check the plugin’s hooks and MCP servers for that. Claude Code’s own memory is a different thing: CLAUDE.md files you write, which Claude reads at the start of a session. See Claude Code memory.

For Hindsight, the supported route for Claude Code is not a plugin:

  • Hindsight’s Coding Agents package is one npm install, npx @vectorize-io/hindsight-coding-agents install claude-code, and it covers 22 CLI agents. It captures memory automatically, with one memory bank per repository. Its docs (read 2026-11-12) say the old per-agent Claude Code plugin is retired and gets no updates.
  • Hindsight also has an MCP server, covered in Claude Code MCP. We did not test that URL in Claude Code.
  • Limit: Hindsight returns more context per retrieval than Mem0 by default, and each save calls an LLM. Set max_tokens lower if the recall is too large.
  • Benchmark: Hindsight reports 64.1 on BEAM 10M against Mem0’s 48.6. Both figures come from Mem0’s own comparison page, not from an independent test.

Sources

  • Claude Code plugin docs, code.claude.com/docs/en/plugins, /plugins/install and the plugin command reference (all read 2026-11-14): what a plugin is, install steps, scopes, commands, context cost.
  • Hindsight Coding Agents docs, hindsight.vectorize.io/sdks/integrations/coding-agents (read 2026-11-12): the install command, 22 agents, the retired Claude Code plugin.
  • Benchmark figures: Mem0’s comparison page, not Hindsight’s own report.