Claude Cowork Plugins: What They Are and Which to Install

Anthropic added plugins to Cowork on January 30, 2026, opening with 11 plugins built from its own team's work (Anthropic). The catalog has grown fast since then. On November 12, 2026, the marketplace file in Anthropic's knowledge-work plugins repository listed 123 plugins, and its official Claude Code marketplace listed 317. That's a lot to sort through when most people need two or three.

This guide explains what a plugin actually contains, how plugins behave differently in Cowork, chat and Claude Code, how to install and manage them, and which ones are worth adding first for each kind of work. All product details come from Anthropic's Help Center, Claude docs and Claude Code docs, read on November 12, 2026.

What are Claude Cowork plugins?

A plugin is a package that turns Claude into a specialist for a job. Anthropic's Help Center describes it as a bundle of "skills, connectors, and sub-agents into a single package, so you get a ready-to-go setup from your first conversation instead of configuring each piece yourself" (Claude Help Center).

In practice, a sales plugin might connect Claude to your CRM and knowledge base, teach it your sales process, and give you commands for prospect research and call follow-ups. That's Anthropic's own example from the launch post. You define what goes in once, and Claude pulls from it whenever the task calls for it (Anthropic).

Plugins are available on all paid plans: Pro, Max, Ultra, Team and Enterprise. Free accounts don't get them (Claude Help Center). If you're new to Cowork itself, our guide to using Claude Cowork covers the basics first.

What's inside a plugin

A plugin can contain four kinds of parts. Not every app loads all four, which is why the same plugin can do more in Cowork than in a chat.

Source: Claude docs: Plugins. Claude Code plugins can also carry hooks and language servers (Claude Code docs).

Under the hood, it's all files. Anthropic's knowledge-work plugins each have a manifest, an .mcp.json file for tool connections, a commands/ folder and a skills/ folder. The repository describes every part as "markdown and JSON, no code, no infrastructure, no build steps" (anthropics/knowledge-work-plugins). That matters later, because it means anyone on your team can read and edit what a plugin tells Claude to do.

One detail people miss: adding a plugin doesn't connect anything for you. Each bundled connector shows as Connected, Not connected or Not added on the plugin's Connectors tab, and a skill that needs a service can't reach it until that connector is connected (Claude docs).

Where plugins work: Cowork, chat and Claude Code

Plugins are saved to your account, not to your computer. Add one in Claude on the web or the desktop app and it follows you (Claude docs):

• Cowork: skills, commands, agents and connectors load into your tasks the next time you start one. Type /plugin-name:command to run a command, or describe the task and let Claude load the matching skill.
• Chat (web, desktop and mobile): skills, commands and connectors are available. Hooks and sub-agents don't run in chat, so they appear grayed out (Claude Help Center).
• Claude Code: the plugin downloads as a synced plugin the next time you start a session signed in to the same account. Run /reload-plugins to load it, then type /plugin-name:skill-name. In Claude Code a plugin runs in full on your computer, including its MCP servers.

The sync only goes one way. Plugins you install from the Claude Code command line stay on that machine and aren't added to your claude.ai account (Claude docs). If you split your time between Cowork and Claude Code, add plugins from Customize so both sides get them. Our Cowork vs Codex guide covers how Cowork compares with coding agents more broadly.

How to install a Cowork plugin

From the directory

1. Open the Cowork tab, then Customize in the left sidebar.
2. Open the Plugins tab, then Discover.
3. Select a plugin to see what it contains, then click Add.
4. Open the plugin's Connectors tab and connect anything it needs.

Discover shows plugins from Anthropic's default marketplaces, the directory, and any your organization provides. Customize also holds your skills and connectors, so it's one place to manage everything (Claude Help Center).

From a marketplace or a file

Under Add, you can choose Add marketplace and enter a Git repository (for example the owner/repo shorthand for GitHub), or Upload plugin to add a folder as a .zip or .plugin file. On Team and Enterprise, colleagues can share a plugin with you directly, and it appears under Shared with you, turned off until you switch it on (Claude docs).

From the Claude Code command line

In a Claude Code session, run /plugin to browse, or /plugin install commit-commands@claude-plugins-official to open a specific plugin's details. Claude Code shows what it will install and asks you to pick a scope (just you, this repository for everyone, or you in this repository only) before anything installs. To use Anthropic's knowledge-work set in Claude Code, add that marketplace first with claude plugin marketplace add anthropics/knowledge-work-plugins (Claude Code docs, GitHub).

To pause a plugin, open it and turn off its Disable plugin toggle. To take it off your account, select Remove from its menu. Plugins from a marketplace or the directory update from their source automatically (Claude docs).

Which Claude Cowork plugins to install

Anthropic's own advice is to try a plugin on a real task and turn it off if it doesn't help (Claude docs). So don't install ten. Pick the one that matches the work you do most, use it for a week, and then decide on the next. These are the knowledge-work plugins Anthropic maintains, with the connectors each one is built around (anthropics/knowledge-work-plugins):

The same marketplace now also carries Engineering, Human resources, Design, Operations and Small business plugins, plus vendor-built plugins such as Slack, Apollo, Canva, Dropbox, monday.com, Intuit QuickBooks and HubSpot Sales. A vendor plugin is the better choice when one tool runs most of your day, since its maker wrote the skills around its own product. Check what each contains on its page before adding it.

A few practical rules for choosing:

• Match the connectors to your stack. A plugin built around HubSpot does less if your team runs Salesforce. You can swap connectors by editing its .mcp.json, and Anthropic encourages it (GitHub).
• Use commands for repeat work. Commands like /sales:call-prep or /finance:reconciliation are worth learning for tasks you do every week. For one-off work, just describe the task.
• Watch how many you stack. Each plugin adds skills Claude has to consider. In Claude Code, plugins from Anthropic's official marketplace show a context cost estimate before install (Claude Code docs). In Cowork, fewer well-fitted plugins beat many generic ones.

If you want ideas for what to point a plugin at, our list of Claude Cowork use cases covers 25 workflows teams run weekly.

Claude Code plugins worth installing

If you came here searching for Claude Code plugins, the picture is a little different. Claude Code adds Anthropic's official marketplace, claude-plugins-official, the first time you start an interactive session. A separate community marketplace, claude-community, carries third-party plugins submitted to Anthropic, and you add it yourself (Claude Code docs). On November 12, 2026, Claude Marketplace showed 529 plugins that work with Claude Code (Claude Marketplace).

These are among the first plugins Claude Marketplace lists, all marked Anthropic verified, grouped by what they're for (Claude Marketplace, claude-plugins-official):

• Writing and reviewing code: Code Review (multiple review agents with confidence-based scoring), PR Review Toolkit, Code Simplifier, Feature Dev and Security Guidance.
• Front end: Frontend Design, for distinctive, production-grade interfaces rather than generic AI layouts.
• Working habits: Superpowers (brainstorming, TDD and systematic debugging), Commit Commands (commit, push and PR creation) and CLAUDE.md Management.
• Tools and docs: Context7 for up-to-date library documentation, Playwright for browser testing, GitHub, and language servers like TypeScript LSP and Pyright LSP.
• Building your own: Skill Creator and Plugin Dev.

Pair these with the skills in our best Claude skills ranking. If you're weighing Claude Code against OpenAI's agent, see Codex vs Claude Code.

Plugins vs skills vs connectors

The three overlap, because a plugin is built from the other two. The simplest way to keep them straight:

For the longer version, including where Projects fit, read Claude Projects vs Skills vs Connectors, browse the Claude connectors list, or see what agent skills actually do. Connectors run on MCP. If you're curious how that differs from a normal integration, MCP vs API explains it.

Customize a plugin or build your own

The knowledge-work plugins are honest about what they are. Anthropic calls them "generic starting points" that "become much more useful when you customize them for how your company actually works" (anthropics/knowledge-work-plugins). There are three ways to do that:

1. Customize in Cowork. Open an installed plugin and click Customize. Cowork opens a task where Claude helps you adjust its skills and connectors to how you work (Claude Help Center).
2. Create with Claude. Go to Customize > Plugins > Add > Create with Claude and describe the plugin you want. A sentence is enough. Claude asks about your workflow, tools, standards and edge cases, then shows the plugin as a file card you save to your account (Claude Academy).
3. Edit the files. Use Create a plugin to write the files directly, or fork a plugin repository and change the markdown: your terminology, your process, your connectors.

On Team and Enterprise plans you can share a plugin with specific people or publish it to your organization, and owners can make a plugin installed by default or required. A plugin's page also shows adoption and runs over the last 30 days (Claude docs). You can submit a plugin to Anthropic's directory too, where it's reviewed before listing.

Whichever route you take, the hard part isn't the file format. It's writing down how good work actually gets done in a way Claude can follow. That's the same problem behind prompt debt and the Brief-Once Workflow.

Before you install: cost, safety and limits

• Cost. Plugins have no separate price on paid plans. They draw on normal usage, and multi-step Cowork tasks that use connected apps use more of it (Claude Help Center). Our Claude Cowork pricing breakdown covers what that means per task. The service behind a connector may need its own account or paid plan.
• Review. Anthropic reviews plugins listed in the directory: every version gets automated validation and a security scan, and a person reviews a new listing before it goes live. That review doesn't cover a plugin you add from a marketplace URL or upload yourself (Claude docs).
• What runs locally. In Cowork and Claude Code, agents and hooks can run commands on your computer, and a connector marked "Runs in each session" runs a program locally. Read a plugin's contents before you add it.
• Desktop-only cases. Plugins that include local MCP servers work through the desktop app only (Claude Help Center).
• Custom connectors must be public. In Cowork, connectors reach services through Anthropic's cloud, so a custom connector has to point to a server reachable over the public internet (Claude Help Center).
• Enterprise scanning. If your Enterprise organization turns on skill scanning, plugins are checked for malicious content when installed or updated (Claude Help Center).

The plugin for expertise: where Callstand fits

A plugin gives Claude a role's process and the tools to carry it out. What it can't give Claude is a particular person's judgment: which of the five reasonable options your best account executive would actually pick, what your head of content would cut from a draft, what a senior ecommerce marketer would tell you to skip. Generic skills produce generic-good work. That's the gap behind the Judgment Layer idea.

Callstand is built for that gap. It isn't a packaged plugin in Anthropic's directory. It connects to Claude as a connector, with one link, and sits in Customize next to your plugins. In Claude Code, Cursor and Codex you paste the same link plus your API key into the app's MCP settings. Here's what it holds and does, according to Callstand's How it works page on November 12, 2026:

• Your know-how, not a template. You describe up to five areas you know well, then calibrate: Callstand writes realistic requests, an AI answers them, and you write what's off and what you'd do instead.
• Three kinds of help. Plan it (a prioritized plan), make it (the finished piece) and check it (what to fix, what's fine and what to stop worrying about). Your AI picks the right one for what you asked.
• Other people's expertise. Connect a teammate's or an expert's Callstand next to yours. They approve your access, and their know-how shows up in your AI. People get answers, never the owner's notes, examples or calibrations.
• Simple pricing. $0.50 per call from prepaid credits, $5 to start, with building free up to 60 calibration test cases and 10 side-by-side tests.

The two work together. A Marketing plugin gives Claude your campaign workflow and connects it to Canva and HubSpot. A Callstand connected beside it lets Claude check that campaign brief against how your most senior marketer would judge it before anything gets built. The plugin handles the steps. The Callstand handles the calls.

Frequently asked questions