How to Install and Use Anthropic Knowledge Work Plugins: A Complete Guide

Anthropic Knowledge Work Plugins extend Claude with domain-specific skills, slash-commands, and MCP connectors that you install via the Claude Cowork marketplace or the Claude Code CLI, requiring no build steps as they rely on declarative JSON and Markdown files.

The anthropics/knowledge-work-plugins repository provides ready-made plugins for roles like sales, legal, and productivity. Each plugin transforms Claude into a specialized assistant by automatically loading contextual knowledge and exposing explicit commands for complex workflows, all without compiling code.

Understanding the Plugin Structure

Every plugin follows a strict directory layout using pure data files. When installed, Claude interprets these files to extend its capabilities immediately.

A typical plugin directory contains:

plugin-name/
├── .claude-plugin/plugin.json   # Manifest defining name and version

├── .mcp.json                    # MCP connector definitions for external tools

├── commands/                    # Slash-commands (e.g., /sales:call-prep)

└── skills/                      # Markdown files encoding domain knowledge

Key components include:

• .claude-plugin/plugin.json: The manifest that tells Claude the plugin's metadata and version.
• .mcp.json: Maps connector names like slack or asana to MCP server endpoints and OAuth settings. Claude contacts these endpoints at runtime to fetch or push data.
• skills/: Contains Markdown files with step-by-step instructions and workflows. Claude automatically triggers these when detecting matching user intent.
• commands/: Houses explicit entry points that you invoke with slash commands followed by optional flags.

Installation Methods

You have two primary paths to install Anthropic Knowledge Work Plugins, depending on whether you use the web interface or terminal tools.

Installing via Claude Cowork

The simplest method uses the graphical marketplace:

1. Navigate to https://claude.com/plugins.
2. Browse available plugins (e.g., productivity, sales, legal).
3. Click Install on your chosen plugin.

The plugin immediately becomes available in your Cowork environment and is ready for use.

Installing via Claude Code (CLI)

For terminal-based workflows, use the Claude Code CLI to add the marketplace and install specific plugins:

# Add the Knowledge Work Plugins marketplace

claude plugin marketplace add anthropics/knowledge-work-plugins

# Install the sales plugin

claude plugin install sales@knowledge-work-plugins

After installation, slash commands become available in any Claude Code session. For example, the productivity plugin exposes:

# Quick update of recent activity

/productivity:update

# Deep scan across chat, email, calendar, and docs

/productivity:update --comprehensive

How Skills and Commands Work

The plugin system distinguishes between automatic skills and explicit commands.

Skills are Markdown files located in directories like productivity/skills/update/SKILL.md. These files contain step-by-step instructions, argument hints, and usage examples. According to the repository source code, Claude automatically loads these skills whenever relevant context appears in your conversation.

Commands provide explicit control through slash invocation. For instance, the sales plugin exposes /sales:call-prep, which accepts specific flags:

# Generate a call preparation brief for a specific account

/sales:call-prep --account AcmeCorp

The complete definition for the Update command resides in productivity/skills/update/SKILL.md, documenting both basic syntax and the --comprehensive flag for deep scanning across integrated tools.

Customizing Plugins for Your Environment

Because Knowledge Work Plugins use declarative configuration, you can customize them without recompiling. Changes take effect immediately upon reinstall or reload.

To customize a plugin:

1. Edit .mcp.json to point to your own tool instances or modify OAuth settings for connectors.
2. Modify skill Markdown files in the skills/ directory to add company-specific terminology or adjust workflows.
3. Create new commands by adding Markdown files to the commands/ directory following the existing layout.

Key Source Files in the Repository

Understanding these critical files helps you navigate the anthropics/knowledge-work-plugins repository:

• README.md (root): Overview, installation instructions, and marketplace information.
• productivity/.mcp.json: Connector definitions for external SaaS tools like Slack and Notion.
• productivity/skills/update/SKILL.md: Example skill showing command syntax and the --comprehensive flag usage.
• sales/README.md: Documentation of the sales plugin’s capabilities and connected tools.
• productivity/CONNECTORS.md: Detailed reference of supported MCP connectors.

Summary

• Install Anthropic Knowledge Work Plugins via the Claude Cowork marketplace or Claude Code CLI using claude plugin install.
• Plugins consist of declarative JSON manifests (.claude-plugin/plugin.json), MCP connector configs (.mcp.json), and Markdown skills/commands.
• Skills auto-load based on context, while Commands require explicit slash invocation (e.g., /productivity:update).
• Customize plugins by editing .mcp.json for connector endpoints or modifying files in skills/ and commands/.
• No compilation is required; configuration changes are effective immediately upon reload.

Frequently Asked Questions

Do I need to compile or build the plugins after editing files?

No. Anthropic Knowledge Work Plugins use a declarative architecture where .mcp.json files and Markdown skills are pure data. When you modify these files and reload the plugin, Claude interprets the updated configuration immediately without a build step or binary compilation.

Can I use Knowledge Work Plugins with the standard Claude API?

These plugins are designed specifically for Claude Cowork and Claude Code environments. While the underlying MCP connectors use standard protocols, the automatic skill loading and slash-command infrastructure require the runtime provided by these Anthropic interfaces rather than the standard API.

How do connectors in .mcp.json authenticate with external tools?

The .mcp.json file maps connector names—such as slack, asana, or notion—to MCP server endpoints and optional OAuth settings. Claude contacts these endpoints at runtime to fetch or push data, handling authentication through the configured OAuth flows without requiring code changes in the plugin itself.

What is the difference between skills and commands?

Skills are Markdown files in the skills/ directory that Claude automatically triggers when it detects matching user intent, providing contextual assistance based on the source content. Commands are explicit entry points in the commands/ directory that you must invoke with a slash prefix (e.g., /sales:call-prep), often accepting flags like --account or --comprehensive for targeted operations.

{
  "mcpServers": {
    "instagit": {
      "command": "npx",
      "args": ["-y", "instagit@latest"]
    }
  }
}