Knowledge Work Plugins Use Cases: Domain-Specific AI Assistants for Business Teams

The Anthropics Knowledge Work Plugins transform Claude into specialized business assistants for productivity, sales, data analysis, legal review, and scientific research through declarative, file-based configurations that require no compilation or runtime services.

The anthropics/knowledge-work-plugins repository provides a collection of ready-to-use extensions that turn Claude (via Cowork or Claude Code) into a domain-specific assistant across diverse business functions. Each plugin follows a uniform architecture—featuring a manifest in .claude-plugin/plugin.json, connector definitions in .mcp.json, Markdown-based skills, and optional slash commands—allowing teams to deploy role-aware AI support instantly without writing code.

Architecture: How Knowledge Work Plugins Function

Before exploring specific scenarios, understanding the plugin structure clarifies how these use cases are delivered. Each plugin is a pure-file specification that Claude discovers and interprets at load time.

File-Based Manifest System

Every plugin root contains a .claude-plugin/plugin.json file that declares the plugin name, version, and exposed commands. Claude automatically discovers these manifests and loads the associated capabilities:

// Located at: <plugin>/.claude-plugin/plugin.json
{
  "name": "productivity",
  "version": "1.0.0",
  "commands": [
    {
      "name": "start",
      "description": "Initialize task tracking workspace"
    }
  ]
}

Connector Abstraction via MCP

The .mcp.json file maps logical connector names (e.g., "CRM", "Data Warehouse") to actual MCP server endpoints. When present, skills can call external APIs like Salesforce, Snowflake, or Jira; otherwise, the skill falls back to "stand-alone" mode using user-provided context.

Skills as Domain Knowledge

Skills reside as Markdown files under the skills/ directory, encoding domain expertise, best-practice checklists, and step-by-step workflows. Claude automatically retrieves relevant skills when user requests align with the skill's description metadata.

Slash Commands for Explicit Actions

Commands live under commands/ and implement deterministic actions (e.g., /forecast or /call-prep). These can be invoked directly in the UI and optionally leverage connectors for enriched, data-driven results.

Productivity and Workplace Memory

Individual contributors and teams use this plugin to maintain persistent, AI-augmented task tracking that syncs with existing tools.

The productivity plugin creates a persistent workspace through TASKS.md and visual dashboards. When you invoke /start, Claude initializes the workspace structure. Later, natural language requests like "I need to review the budget proposal for Sarah by Friday" automatically generate tasks with context-aware due dates and owners.

Key files in anthropics/knowledge-work-plugins/tree/main/productivity/:

• productivity/README.md – Overview, available commands, and skill definitions
• productivity/.mcp.json – Connector definitions for Slack, Asana, or Jira integration
• productivity/skills/ – Task management methodologies and prioritization frameworks

### Example: Natural Task Creation

You: I need to review the budget proposal for Sarah by Friday, and draft the Q2 roadmap after syncing with Greg.

Claude: 
- Updates `TASKS.md` with two entries annotated with due dates and owners
- Refreshes `dashboard.html` to display the new tasks
- Asks: "Do you want me to schedule reminders in your calendar?"

Sales and Revenue Operations

Sales representatives leverage this plugin for prospect research, call preparation, pipeline forecasting, and asset generation.

The sales plugin integrates with CRM systems through MCP connectors to provide data-driven insights. The /call-prep command assembles briefing documents by pulling account data, recent news, and key contacts. The /call-summary command extracts action items from transcripts and drafts follow-up emails.

The account-research skill triggers automatically when users request pre-call intelligence, scraping relevant data and generating competitive positioning briefs.

Data Analysis and Business Intelligence

Data teams utilize this plugin for SQL authoring, data exploration, visualization, and dashboard building without boilerplate code.

When connected to warehouses like Snowflake, BigQuery, or Databricks via .mcp.json configurations, Claude executes queries directly. In stand-alone mode, users upload CSVs for immediate analysis. The /analyze command interprets natural language questions, generates SQL or Python code, executes the analysis, and returns interactive Chart.js visualizations.

### Example: Ad-hoc Analytics

You: /analyze What was our monthly revenue trend for the past 12 months?

Claude:
1. Writes SQL query optimized for your configured warehouse
2. Executes against the connected database via MCP connector
3. Generates `dashboard.html` with interactive charts
4. Provides strategic insights on detected anomalies

Customer Support and Success

Support agents streamline ticket triage, response drafting, and knowledge-base generation.

Agents paste ticket text or transcripts to receive structured response drafts, suggested next steps, and optional knowledge-base article generation. The plugin enforces brand voice standards through skills that capture tone guidelines and resolution workflows, reducing response time while maintaining quality consistency.

Product Management

Product managers synthesize user research, draft technical specifications, and maintain roadmap documentation.

The plugin pulls data from Jira, Notion, or Figma via MCP connectors to compare features across competitors or generate stakeholder updates. Skills encode frameworks for prioritization (RICE, MoSCoW) and spec writing templates, allowing Claude to draft implementation documents that reference existing codebase patterns and design files.

Marketing and Content Operations

Marketing teams generate campaign briefs, draft copy, and create performance dashboards.

The plugin enforces brand voice through skills that contain style guides and messaging hierarchies. When connected to HubSpot, Amplitude, or Google Analytics via .mcp.json, the /report command generates performance summaries with visualizations. Content drafting skills help maintain consistency across blog posts, email sequences, and social media assets.

Legal and Compliance

Legal teams accelerate contract review, NDA triage, and compliance verification.

Users feed contract documents to Claude to receive risk summaries, suggested clause edits, and templated responses to counterparties. Skills encode regulatory requirements (GDPR, SOC2) and standard playbooks, allowing for preliminary contract analysis before attorney review. The plugin stores approved language templates in skills/ for consistent drafting.

Finance and Accounting

Finance professionals automate journal entry preparation, account reconciliation, and variance analysis.

When integrated with Snowflake or BigQuery, Claude can reconcile accounts against transactional data, generate trial balances, and produce audit-ready reports. The plugin's skills encode accounting standards and internal controls, ensuring outputs meet compliance requirements. Users generate variance analysis narratives by simply asking Claude to explain period-over-period discrepancies.

Enterprise Search

Knowledge workers query across fragmented information silos using natural language.

This plugin enables employees to issue single queries that search Slack, Notion, Confluence, Google Drive, and other repositories simultaneously. The MCP connectors index these sources (where API access permits), and Claude synthesizes consolidated answers with source citations. This eliminates context switching between applications during research tasks.

Bio-Research and Life Sciences

Researchers accelerate literature reviews, genomics analysis, and target prioritization.

Through integrations with PubMed, Benchling, and bioinformatics platforms, the plugin pulls latest pre-prints on specific targets, runs scRNA-seq analysis pipelines, and generates comparative tables of candidate compounds. Skills encode domain-specific methodologies like CRISPR guide design or pathway analysis workflows.

Custom Plugin Development

Platform teams clone the repository to create organization-specific plugins tailored to internal processes.

The cowork-plugin-management plugin (located in the cowork-plugin-management/ directory) provides templates and validation tools for creating new plugins. Teams edit Markdown skills and connector definitions in .mcp.json to encapsulate proprietary business logic without deploying additional infrastructure.

claude plugins add knowledge-work-plugins/cowork-plugin-management

Summary

• Declarative Architecture: Knowledge work plugins use pure files (JSON manifests, Markdown skills) requiring no compilation or runtime services
• Connector Flexibility: The .mcp.json abstraction allows the same skills to function in connected mode (with live data) or stand-alone mode (with user uploads)
• Domain Coverage: Plugins span Productivity, Sales, Data Analysis, Customer Support, Product Management, Marketing, Legal, Finance, Enterprise Search, and Bio-Research
• Extensibility: Organizations can author custom plugins by cloning the repository and editing skill files in the skills/ directory
• Zero-Code Deployment: Installation via claude plugins add immediately exposes slash commands and domain expertise without API development

Frequently Asked Questions

How do I install a knowledge work plugin?

Navigate to your Cowork or Claude Code environment and run the installation command for the specific plugin you need. For example, claude plugins add knowledge-work-plugins/sales installs the sales plugin. Claude automatically discovers the .claude-plugin/plugin.json manifest and loads the associated skills and commands located in the skills/ and commands/ directories.

Can knowledge work plugins access my company's internal databases?

Yes, when you configure the .mcp.json file in the plugin root with valid MCP server endpoints. This file maps logical connector names (like "CRM" or "Data Warehouse") to actual API connections. If connectors are not configured, plugins fall back to stand-alone mode and operate on files you manually upload or paste into the conversation.

What is the difference between a skill and a slash command?

Skills are Markdown files under the skills/ directory containing domain expertise, best practices, and workflows that Claude retrieves automatically when relevant. Slash commands are explicit actions (like /call-prep or /analyze) located in the commands/ directory that users invoke directly to trigger specific, deterministic workflows. Skills provide passive knowledge; commands provide active functionality.

How do I create a custom plugin for my internal team?

Clone the anthropics/knowledge-work-plugins repository and use the cowork-plugin-management plugin as a template. Create a new directory with a .claude-plugin/plugin.json manifest, add your domain knowledge as Markdown files in skills/, define any API connections in .mcp.json, and place workflow commands in commands/. Your team can then install the custom plugin directly from your repository URL.

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