A plugin is an installable bundle that adds several abilities to Claude in one step.

Anthropic describes a plugin as a package that extends what Claude can do. Installing one can add skills, connectors to outside tools, slash commands, helper agents, and other pieces together, instead of setting each one up by hand. Plugins are usually built around a job function, such as sales, finance, legal, data analysis, or operations, so that Claude has a strong starting point for helping anyone in that role.

Plugins first became prominent in Claude Cowork, the desktop experience for longer, hands-on tasks, and have since appeared in other places too. Because this area is still developing, the exact names, locations, and availability can change. This guide explains the ideas, which are stable, and points you to the current help center for the details that are not.

These terms are easy to mix up, so here is a clear comparison.

Skills, connectors, and slash commands can each exist on their own. A plugin is the container that bundles them for a purpose. Understanding the difference helps you decide whether you need a whole plugin or just one piece.

flowchart TD
    P["Plugin
one installable package"] --> S["Skills
how to do the work"]
    P --> C["Connectors
access to your tools"]
    P --> M["Slash commands
shortcuts you trigger"]
    P --> A["Sub-agents
specialist helpers"]
    P --> H["Hooks
automatic actions"]
        

Each part plays a different role, and knowing them helps you use and adjust a plugin.

They save setup time and make results more consistent.

Without a plugin, every person on a team may explain the same process to Claude in a slightly different way, and get slightly different results. A plugin captures the preferred way of working once, so everyone starts from the same standard. It also saves you from rebuilding the same setup in every new conversation.

Plugins are not limited to a single screen, but not every part works everywhere.

Plugins were introduced with Claude Cowork and are managed from a customization area in the Claude apps. Anthropic’s help center for organizations says plugins distributed by an organization appear in both regular chat, on the web and in the Claude Desktop chat tab, and in Cowork. Claude Code, the coding tool, has its own plugin system too, with its own way of adding plugin collections.

Third-party guides summarizing Anthropic’s documentation note that skills and commands work across those places, while hooks and sub-agents run only in Cowork and Claude Code, and may appear greyed out in chat. Behavior and availability change, so check the current help center to see what your plan and app support.

There are several sources, and each has a different level of trust.

Plugins can come from the marketplace or directory that Anthropic runs, from your own organization, or from a file someone gives you. Anthropic provides plugins built for common functions, such as sales, finance, legal, data, and enterprise search, and its directory lists plugins and connectors people can browse inside Claude. Anthropic says it checks directory submissions before listing them. Organizations can also publish their own curated collections, and individuals can upload a plugin package they built or received.

Installation is usually just a few clicks, but a little checking first is wise.

According to Anthropic’s help center for Cowork, you open the Claude Desktop app, go to the Cowork area, open the customization menu in the sidebar where your plugins, skills, and connectors live, and choose to browse plugins. You pick a plugin and click install. You can also upload a plugin file if you built one or received one from a colleague, and plugins you add yourself are saved locally on your machine. Menu names can change, so follow the current instructions.

Commands are shortcuts you trigger on purpose.

Once a plugin is installed, it adds commands you can use in your work. In Cowork, typing a slash or using the add button shows the commands available from your installed plugins. A command usually starts a particular workflow, such as preparing for a sales call, writing a data query, or reviewing a contract. Some commands show a short form to fill in, so that Claude gets the details it needs.

Commands are useful when you know exactly what you want. Skills, by contrast, tend to activate by themselves when Claude judges that they fit. Using both gives you a mix of control and convenience.

Much of a plugin’s value works quietly in the background.

Skills inside a plugin hold the knowledge and standards for a domain, such as how your team formats a report, which checks to run on a dataset, or what a good brief looks like. Claude draws on them when a task fits, so you do not have to remember to ask. This is why two people with the same plugin tend to get more similar results than two people prompting from scratch.

You can usually tell a skill was used from how the answer is structured, or because Claude mentions it. If a skill is not firing when you expect, name the task more clearly, or mention the skill by name. If a skill fires when it should not, you can turn that component off or adjust it.

Connectors are the part that touches your real data, so they deserve the most thought.

Many plugins include connectors to tools such as email, documents, customer systems, chat apps, or data warehouses. When you authorize a connector, Claude can read from that tool, and sometimes act in it. This is what makes a plugin practical, since the work happens on your real information. It is also where privacy and safety matter most.

Connect only what the plugin needs for your task. Read what each connector will be allowed to do. If a plugin asks for access that seems unrelated to its purpose, pause and ask why. On work accounts, your administrator may control which connectors are allowed, so ask before connecting anything sensitive.

A plugin is a starting point, and the best ones are adjusted to your own way of working.

Anthropic’s help center describes a way to customize a plugin by starting a conversation with Claude that asks it to adjust the plugin’s skills, commands, and connectors to match how you work. You can tell Claude about your templates, your terminology, your approval steps, and the tools you actually use. This turns a general plugin into one that fits your team.

Make changes in small steps and test after each one. Write down what you changed and why, so you can undo it if something stops working. If you edit a plugin that other people also use, tell them what changed.

You do not have to accept a plugin as an all-or-nothing package.

Anthropic’s guidance says that after installing a plugin, you can open it and enable or disable individual components. This is a useful safety and focus tool. If you only want the skills and not the connectors, you can leave the connectors off. If a command clutters your list, you can switch it off. Keeping only what you use makes Claude’s behavior easier to understand.

When no ready-made plugin fits, you can package your own process.

Anthropic offers a plugin whose purpose is to help you build custom plugins from scratch. The usual idea is to describe a process you repeat, let Claude help turn it into skills and commands, connect the tools it needs, test it, and then share it with your team. You do not need to write code to begin, because the central work is clearly describing how a task should be done.

Before building, check whether an existing plugin already does most of what you need. It is often faster to customize something that exists than to create something new. When you do build, start with one workflow, make it reliable, and expand slowly.

flowchart LR
    A["Pick one repeated process"] --> B["Describe how it should be done"]
    B --> C["Turn it into skills and commands"]
    C --> D["Add only the needed connectors"]
    D --> E["Test on real examples"]
    E --> F{"Reliable?"}
    F -- "No" --> B
    F -- "Yes" --> G["Share with your team"]
        

The biggest payoff comes when a whole group works from the same setup.

Anthropic’s help center describes plugin marketplaces that let Team and Enterprise owners distribute curated plugins to everyone in their organization. An owner creates a marketplace, adds plugins to it, and controls which plugins members can see and use. Anthropic also provides built-in marketplaces for functions such as legal and finance, and a general knowledge work marketplace may be added by default. Plugins distributed this way appear in both chat and Cowork.

Organizations can also publish a plugin in Anthropic’s public directory, which involves a review. If you only want to share inside your company, using your own marketplace is the more appropriate route.

At work, the list of plugins you see may be shaped by your organization.

According to Anthropic’s help center, administrators can decide how each plugin appears to members. A plugin can be offered for self-service installation, installed automatically, required, marked not available, or hidden from the catalog entirely, which is useful for staging or retiring a plugin. If a plugin is missing or cannot be removed, an organization setting may be why.

A plugin can instruct Claude and connect it to your tools, so trust matters.

Because a plugin bundles instructions, tool connections, and sometimes automatic actions, installing one is like giving a new colleague a set of instructions and some keys. Only install plugins from sources you trust, such as Anthropic’s directory, your organization, or people you know. Be careful with plugin files from unknown senders. Some guidance for Cowork specifically advises installing connectors only from trusted sources.

Read what a plugin does before you rely on it. Look at its skills for odd instructions, check which connectors it asks for, and be extra careful with hooks, since they can run automatically. Remember that text inside documents and web pages is information for Claude to read, not an order for it to follow, and a trustworthy plugin should not ask you to hide what it is doing.

Familiar jobs show how plugins fit into a working week.

Plugins save effort, and they ask for attention to access and upkeep.

These habits cause most plugin problems.

Follow these steps for almost any plugin.

A few popular beliefs about plugins are only partly true.

Quick answers to the questions beginners ask most.

It is an installable package that bundles skills, connectors, slash commands, and sometimes sub-agents and hooks so Claude can work like a specialist for a job function.

A skill is a single set of reusable instructions. A plugin bundles several components, including skills, into one installable package.

Look in the customization area of the Claude app, where you can browse plugins or upload a plugin file. Exact menu names change, so follow the current help center.

Anthropic says plugins distributed by an organization can appear in chat and in Cowork. Some components, such as sub-agents and hooks, may be limited to certain environments.

Yes. You can ask Claude to adjust its skills, commands, and connectors to match how you work, and you can switch individual components on or off.

Yes. Anthropic offers a plugin that helps create custom plugins. Start with one repeated process, test it on real examples, and then share it.

They are as safe as their source and what they connect to. Use trusted sources, read the contents, grant the least access, and keep approvals on for actions.

An administrator may have marked it unavailable or hidden it, or your plan may not include it. Ask your administrator and check the help center.

Team and Enterprise owners can create a marketplace, add plugins, and control which ones members can see and install.

No. Install only those that solve real tasks, trim the components you do not need, and remove unused ones regularly.

Yes. A plugin improves the starting point, but you should verify important facts and keep a person responsible for decisions.

A plugin is a package that gives Claude a ready-made setup for a job function by bundling skills, connectors, slash commands, and sometimes sub-agents and hooks. It saves setup time, improves consistency, and makes it easy to share good ways of working. You can find plugins in Anthropic’s directory, in your organization’s marketplace, or from a file, install them from the customization area, switch components on and off, and customize them in small steps. Because plugins carry instructions and tool connections, trust matters: prefer reviewed and known sources, grant only the access you need, keep approvals on for actions, and verify the results. Features and availability are still changing, so check the current help center.