Claude Plugins: How to Find, Install, and Use Plugins, Connectors, and Skills

Only one of Claude's three extension types works on the free plan, and it's not "plugins." Skills are free on every tier. Plugins require a paid subscription (Pro, Max, Team, or Enterprise), and connectors limit free users to a single custom integration, according to Anthropic's Help Center. Most "claude plugins" guides ignore these distinctions entirely, treating all three as interchangeable. The result is readers who install the wrong type and hit permission walls they didn't expect.

What each type does

Plugins are pre-configured packages that bundle skills, connectors, and sub-agents into a single install. Think of them as role-specific starter kits. Anthropic provides plugins for sales, finance, legal, marketing, HR, engineering, design, operations, and data analysis. When you install a sales plugin, you get CRM-related connectors, outreach skills, and sub-agents that work together without separate configuration. Some organizations also get access to specialized marketplace modules for life sciences and financial services.

Connectors are OAuth-based integrations that let Claude read from and take actions in external services. Slack, Google Drive, Microsoft 365, and Linear are among the available options. Unlike plugins, connectors do one thing: they bridge Claude and an external app. Some connectors are read-only. Others can send messages, create issues, or modify files. Claude inherits each user's permissions from the connected service, so you only access content you'd normally be authorized to view. Certain connectors are interactive and can render live interfaces (like dashboards or task boards) directly in conversations.

Skills are instruction folders that teach Claude specific, repeatable approaches to tasks. Anthropic builds and maintains a set of default skills for tasks like creating documents in Excel, Word, PowerPoint, and PDF formats. These activate automatically when relevant. Partner companies like Notion, Figma, and Atlassian provide professionally built skills that works alongside their respective MCP connectors. You can also create custom skills that encode your own workflows, from brand-specific document formatting to meeting-note templates. Skills use progressive disclosure: Claude reviews available skills when you submit a task, loads the relevant ones, and applies their instructions without cluttering the conversation.

Here's how the three types compare:

Plugins

• Bundled packages of skills, connectors, and sub-agents
• Paid plans only (Pro, Max, Team, Enterprise)
• Install from the Plugins tab in Customize
• Pre-configured for roles like sales, finance, and engineering

Connectors

• OAuth integrations to external services
• All plans (free users get one custom connector)
• Connect from the Connectors tab
• Read data from and take actions in apps like Slack, Google Drive, and Linear

Skills

• Instruction folders for specialized tasks
• All plans including free
• Install from the Skills tab
• Partners include Notion, Figma, and Atlassian

Platform support also differs. Connectors work across Claude's web interface, desktop app, mobile apps (iOS and Android, with installation currently in beta), Claude Code, and the API. Plugins work on the web, Claude Desktop's Chat tab, and Claude Cowork. Hooks and sub-agents inside plugins only function within Claude Cowork. In regular chat, those components appear grayed out.

If you're looking for Claude Code plugins specifically (the developer-focused CLI tool), those use a separate plugin system covered in the installation section below. This taxonomy applies to Claude's web interface, desktop app, and mobile apps.

Anthropic's unified directory puts skills, connectors, and plugins in a single location with separate tabs for each type. You don't need to navigate between different settings panels.

To open the directory:

1. Open Claude on the web or Claude Desktop
2. Click Customize in the left sidebar
3. Select the tab you want: Skills, Connectors, or Plugins
4. Click the + button, then select Browse followed by the item type

Each tab shows available extensions with type-specific action buttons. Skills show an Install button. Connectors show a Connect button that kicks off an OAuth authentication flow. Plugins show an Install button that adds the full bundle to your account.

Skills you install from the directory are view-only. You can use them in conversations, but you can't edit their instructions. If you want to modify a skill's behavior, create a custom version instead.

On Team and Enterprise plans, the directory includes a Your Organization tab. This surfaces skills, connectors, and plugins that administrators have shared across the team. Organization-managed plugins may be auto-installed or required for all members, depending on admin settings.

If a colleague shares a skill directly with you (outside the directory), it bypasses the browsable interface and goes straight to your personal skills list. You'll find it alongside your other installed skills without searching the directory.

Installed skills default to enabled. You can toggle them off while keeping them in your directory, which is useful when you want to narrow Claude's active skill set for a specific project. Organization-shared skills can't be fully removed by individual users. Only the original sharer or an organization owner can delete them.

The directory does not support cross-type search. You can't type "Slack" and see the Slack connector, Slack-related skills, and any plugin that includes Slack access in one combined view. Check each tab separately when you're looking for everything related to a specific service.

When evaluating extensions, pay attention to what the extension actually does versus what its name suggests. A "Slack plugin" and a "Slack connector" are different things. The connector gives Claude direct OAuth access to Slack's API. A plugin that involves Slack bundles that connector with other skills and sub-agents for a broader workflow setup.

For Claude Code users, plugin discovery works differently. Run claude plugin search from the command line to browse available plugins, or visit the anthropics/claude-plugins-official GitHub repository for the curated list. Community plugins are listed in the separate claude-community marketplace, which you can add with the /plugin marketplace add command.

Installing a plugin

Open the Customize menu, go to the Plugins tab, and click Browse plugins. Select the plugin you want and click Install. The plugin's bundled skills immediately appear in your skill list. Access them by typing / in the chat input or clicking the + button.

You can also upload custom plugin files. If a colleague builds a plugin package for your team's workflow, import it directly through the Plugins tab without going through the public directory.

Organization administrators on Team and Enterprise plans can distribute managed plugins that install automatically for all members. These can't be edited or removed by individual users, which ensures consistent tooling across the team.

Connecting a connector Connectors require an OAuth authentication step because they access your data in external services:

1. Open the Connectors tab in the Customize menu
2. Find the service you want (Slack, Google Drive, Linear, Microsoft 365)
3. Click Connect or Install
4. Complete the authentication prompts from the external service
5. Configure permissions if prompted

After authentication, the connector is active. Claude can suggest using connected apps during conversations when the context is relevant, or you can manually enable specific services for individual chats.

On Team and Enterprise plans, administrators enable connectors at the organization level first. After that, each team member still authenticates individually with their own credentials. Administrators can restrict connector actions to read-only, require approval for certain operations, or block specific actions entirely.

Custom connectors must point to servers accessible over the public internet from Anthropic's IP ranges. Private-network or firewalled services won't work without allowlisting.

Adding skills

Skills install in one handoff flow from the Skills tab. Browse the directory, find a skill, and click Install. The skill becomes available in your conversations immediately.

Anthropic's built-in skills for document creation (Excel, Word, PowerPoint, PDF) activate automatically when relevant tasks come up. No manual installation needed.

Custom skills are different. You create them yourself by writing instruction folders that encode your specific workflows: brand style guides, email templates, data analysis procedures, or task-creation flows for project management tools like Jira, Asana, or Linear.

Claude Code plugins

Claude Code has its own plugin system, separate from the web and desktop directory. These plugins target developers and can include skills, agents, hooks, MCP servers, LSP servers, and background monitors.

Install from the CLI:

claude plugin install @official/plugin-name

Anthropic maintains two public marketplaces: claude-plugins-official (curated by Anthropic, with plugins from GitHub, Playwright, Supabase, Figma, Vercel, and others) and claude-community (third-party submissions that pass automated review).

Claude Code plugins use a namespace system to prevent conflicts. A skill from a plugin called my-tool is invoked as /my-tool:skill-name rather than just /skill-name. You can also develop your own plugins and share them through private team marketplaces or the community submission process.

For development workflows that produce files and artifacts, external workspace services provide persistent storage that agents can access through standard protocols. Options range from S3 buckets and Google Cloud Storage to dedicated workspace platforms. Fast.io offers MCP-based workspace access with a free agent plan that includes 50GB of storage, no credit card required. Its MCP server at mcp.fast.io exposes workspace, file, and AI operations through Streamable HTTP.

Store and share what your Claude agents create

Free 50GB workspace with a Streamable HTTP MCP endpoint. Your agents write files, your team reviews them. No credit card required.

Sign Up Free

The right combination of extensions depends on your daily work. Here are concrete setups for common roles.

Engineering teams

Start with the Engineering plugin for a baseline set of code-related skills and sub-agents. Add the GitHub connector for repository access, pull request creation, and code review. The Linear connector handles issue tracking. If your team follows a specific code review process, create a custom skill that encodes your checklist and quality standards so every review follows the same structure.

For development workflows that produce build artifacts, documentation, or test reports, a shared workspace keeps outputs organized and accessible to the full team. Agents build in their environment, write results to a workspace, and team members review and approve deliverables through the same interface.

Sales teams

The Sales plugin bundles CRM-oriented skills and outreach connectors. Add the Slack connector so Claude can draft messages in your sales channels. Connect Google Drive or Microsoft 365 to give Claude access to proposals, contracts, and pitch decks.

Build a custom skill that encodes your pricing tiers, competitive positioning, and objection-handling framework. This helps Claude generate accurate, on-brand responses to prospect questions instead of generic answers.

Legal teams

Install the Legal plugin for compliance-oriented skills. Connect Microsoft 365 or Google Drive for document access. The connector permission model works well for legal use cases: Claude inherits your document permissions, so it can only access files you're authorized to view. The access controls are already in place through your existing document management system.

The real value is in custom skills for contract review, regulatory analysis, or due diligence checklists. A skill that walks Claude through your firm's standard review process produces consistent work product regardless of who runs the analysis.

Marketing and content teams

The Marketing plugin provides content planning and campaign management skills. Connect Notion for editorial calendars and content databases. The Figma partner skill gives Claude access to design files for generating implementation code or reviewing visual assets.

Content teams that produce large volumes of assets typically follow a workflow where Claude generates drafts, a human reviews and approves, and then the final version goes to the publishing platform. This is where workspace services with built-in AI indexing add value. Upload your brand guidelines, past campaigns, and style references to a searchable workspace. Google Drive offers basic file search, but platforms with semantic indexing, like Fast.io's Intelligence Mode, make your content library queryable through natural language. Ask questions about your existing content and get cited answers instead of scrolling through folders.

Choosing your setup

Start with the plugin for your role to get the bundled defaults. Add connectors one at a time as you identify which external services Claude needs direct access to. Build custom skills only after you've used the defaults enough to know which specific workflows need standardization. Over-configuring upfront leads to skill clutter that makes Claude less responsive to simple requests.

Team and Enterprise administrators control which extensions are available organization-wide and how they behave.

Distributing plugins across your organization

Organization owners can push plugins to all team members through managed distribution. Managed plugins install automatically and can't be modified or removed by individual users. This ensures every team member has access to the same tooling. Administrators can also mark plugins as optional, letting individuals decide whether to install them.

Setting connector permissions

Connectors present the biggest security surface since they grant Claude access to external data. Administrators have three levels of control:

• Read-only: Claude can pull data but can't modify anything in the connected service
• Approval-required: Claude drafts actions, a human reviews and approves before execution
• Blocked: Specific actions or entire connectors are unavailable

Even after an administrator enables a connector at the organization level, each team member authenticates individually with their own credentials. Claude can only access content that the specific user has permission to view in the source system.

Building custom skills for your team

Organizations often get the most value from custom skills that encode internal processes. A legal team's contract review checklist, an engineering team's deployment procedure, a marketing team's brand voice guidelines: these are the workflows where standardization produces consistent results.

Organization-provisioned skills appear automatically in all members' skill lists. Administrators set them as enabled or disabled by default. Individual users can toggle organization skills off but can't fully remove them.

Security considerations

Local MCP servers (used by some connectors and Claude Code plugins) install with the same permissions as any other program on your machine. Only install extensions from sources you trust.

Third-party connectors maintain their own data infrastructure and operate under their own terms of service. Data transfers between Claude and connected services use encryption, but the security boundary is the permission model of the connected service itself. Broad permissions in Google Drive mean broad access for Claude.

For teams handling sensitive data, pair read-only connector restrictions with an approval workflow for any write operations. Add a workspace with audit trails and granular permissions for agent outputs. Google Drive and Dropbox provide basic activity logs, while dedicated agent workspace platforms like Fast.io offer file-level permission controls and full audit histories for everything your agents produce.

Shared chat limitations

Connectors are currently unavailable in shared chats on Team and Enterprise plans. If you share a conversation with a colleague, connector-dependent features in that chat won't have live access to external data. Capture connector outputs within the conversation before sharing to preserve the information.