When Anthropic launched Claude Cowork plugins on January 30, 2026, I spent the next 72 hours testing every single one. I fed the Legal plugin actual client contracts. I gave the Finance plugin real P&L statements. I let the Marketing plugin loose on our Q1 campaign planning. Some plugins absolutely blew me away with their capabilities. Others… well, they need work.
This guide shares everything I learned—what each plugin actually does versus what the marketing claims, which ones deliver immediate value, which ones need careful configuration to be useful, and how to build custom plugins when the official ones don’t quite fit your needs.
If you’re just getting started with Cowork and want to understand the fundamentals first, I recommend reading our main Claude Cowork guide which covers what Cowork is and why it matters. For technical setup and security configuration, see our setup and security guide. This guide assumes you have Cowork installed and are ready to dive into actually using it.
Understanding the Plugin Architecture
Before diving into specific plugins, let’s understand how they work. This matters because once you grasp the underlying structure, you can modify existing plugins or build your own.
Plugins are essentially instruction sets that customize how Claude operates within a workspace. Think of base Claude as a highly capable generalist. Plugins turn it into a specialist.
Each plugin contains four key components:
Skills are domain knowledge that Claude automatically draws on. The Legal plugin’s skills include contract law principles, common clauses, risk assessment frameworks, and industry-standard legal terminology. When you ask a legal question, Claude references these skills without you needing to prompt it explicitly.
Slash Commands are shortcuts for specific tasks. Instead of typing out a detailed request like “Please review this contract for liability issues and flag concerning clauses,” the Legal plugin adds a /review-contract command that does this automatically. Commands streamline repetitive workflows.
MCP Connectors (Model Context Protocol) integrate with external tools. The Sales plugin can connect to your CRM (Salesforce, HubSpot), pulling customer data and interaction history directly into Claude’s context. This means Claude understands your specific sales processes and customer relationships, not just generic sales concepts.
Sub-agents are specialized workers for particular subtasks. The Data Analyst plugin might use sub-agents for data cleaning, visualization creation, and statistical analysis. Each sub-agent is optimized for its specific role, and they work together to handle complex multi-step analyses.
Anthropic open-sourced all eleven official plugins on GitHub, which means you can inspect exactly how they work, modify them for your specific needs, or use them as templates for building entirely custom plugins.
The more you use plugins within a workspace, the more Claude learns about your specific workflows and preferences. This learning is workspace-specific, so your legal workspace becomes increasingly tuned to your legal processes, while your marketing workspace separately learns your marketing voice and strategies.
Deep Dive: The 11 Official Plugins
Let’s go through each official plugin in detail. I’ll cover what it does, how to configure it, real-world use cases I’ve tested, limitations to watch for, and specific tips from hands-on experience.
1. Productivity Plugin
What it does: Task management, calendar coordination, and personal workflow optimization.
Key commands:
/update– Scans emails, calendar, and messages to update your task list/plan-day– Creates a prioritized daily schedule based on tasks and meetings/summarize-inbox– Generates a digest of important emails and action items
Configuration: Connect it to your email (Gmail, Outlook), calendar (Google Calendar, iCal), and task manager (Notion, Todoist, Asana). The plugin uses MCP connectors to access these services.
In my testing, the /update command is genuinely useful. I ran it Monday morning after a busy weekend, and it correctly identified:
- Three urgent emails requiring responses
- Two meetings where I needed to prepare materials
- A deadline I’d forgotten about for Thursday
- Suggested rescheduling a low-priority meeting to free up time for urgent work
The accuracy was around 85%—occasionally it flagged things as urgent that weren’t, but never missed actually important items.
Real-world use case: Daily morning routine. Run /update first thing to get oriented. Use /plan-day to structure your schedule. Throughout the day, use /summarize-inbox before context-switching between projects so you don’t miss important emails.
Limitations: Struggles with complex scheduling preferences. If you have intricate rules about when you’re available or prefer certain meeting types at certain times, you’ll need to provide detailed configuration. Also, it sometimes overschedules—suggesting back-to-back meetings without buffer time.
Pro tip: Create a morning routine prompt that combines commands: “Run /update, then /plan-day considering I have deep work time from 9-11 AM and prefer meetings in the afternoon.” Save this as a custom shortcut.
2. Enterprise Search Plugin
What it does: Searches across all your company tools simultaneously and synthesizes results with citations.
Key commands:
/search [query]– Searches across connected tools/find-all [topic]– Comprehensive search including historical documents/what-did-we-decide [topic]– Surfaces past decisions and rationale
Configuration: Connect to Slack, Google Drive, Notion, email, and any other tools where your company stores information. The more sources connected, the more valuable this plugin becomes.
This might be the most powerful plugin for knowledge workers. The ability to search everything at once is transformative. I tested it with “customer churn analysis” across a workspace connected to:
- Slack (found three relevant channel discussions)
- Google Drive (surfaced two research reports and a presentation)
- Notion (found meeting notes from a customer success review)
- Email (pulled up a thread with a data analyst about churn metrics)
Claude synthesized all of this into a coherent summary with links to each source. What would have taken me 30 minutes of searching and reading happened in 60 seconds.
Real-world use case: Onboarding new employees. When someone joins and asks “where do we document our sales process,” instead of manually finding docs, run /search sales process documentation and Claude finds everything relevant. New hires get comprehensive answers instantly.
Limitations: Requires thorough permission configuration. If Claude can’t access certain Slack channels or folders, search results will be incomplete. Also, it sometimes surfaces older, outdated information without clearly indicating it’s no longer current.
Pro tip: Use /what-did-we-decide before making significant decisions. It prevents re-litigating settled questions and surfaces historical context that might be relevant. “Why did we choose vendor X over vendor Y?” often has important context buried in year-old meeting notes.
3. Marketing Plugin
What it does: Content creation in brand voice, campaign planning, and product launch coordination.
Key commands:
/draft-post [platform] [topic]– Creates social media content/plan-campaign [goal]– Generates comprehensive campaign plans/analyze-performance [campaign]– Reviews metrics and suggests improvements
Configuration: Provide brand guidelines (tone, voice, key messages), access to past successful content for Claude to learn your style, and connections to analytics tools (Google Analytics, social media insights).
I tested this extensively for our Q1 campaign. Initial results were mixed—the content was generic and didn’t match our voice. But after feeding it:
- Ten of our best-performing blog posts
- Social media posts with high engagement
- Our brand voice guidelines
- Examples of content that flopped and why
The quality improved dramatically. The /draft-post command started generating content that genuinely sounded like us. Not perfect, but solid first drafts that needed editing rather than complete rewrites.
Real-world use case: Content calendar production. Use /plan-campaign to outline a month’s worth of content, then /draft-post to create first drafts for each item. What used to take our team two full days now takes about four hours (including editing Claude’s outputs).
Limitations: Struggles with highly creative or conceptual work. If you want something truly original and unexpected, Claude’s output tends toward safe, conventional approaches. It’s better at executing a defined strategy than creating breakthrough creative concepts.
Pro tip: Use it for variation generation. When you have one piece of content that worked well, ask Claude to create variations for different platforms or audiences. “Take this blog post and adapt it for LinkedIn, Twitter, and our email newsletter, adjusting tone and length appropriately.”
4. Sales Plugin
What it does: CRM integration, prospect research, call preparation, and follow-up automation.
Key commands:
/research-prospect [company]– Compiles information on potential customers/call-prep [prospect]– Generates talking points and questions/follow-up [meeting]– Drafts personalized follow-up emails
Configuration: Connect to your CRM (Salesforce, HubSpot, Pipedrive), company knowledge base, and case study repository. Teach Claude your sales methodology and qualification criteria.
The /research-prospect command is impressive. I gave it “Acme Corporation” and it:
- Summarized the company (industry, size, revenue, recent news)
- Identified key decision-makers and their backgrounds
- Found relevant case studies from similar companies we’ve worked with
- Suggested talking points based on their likely pain points
- Flagged potential objections based on their industry
This level of preparation used to require 30-45 minutes of manual research. Claude did it in under two minutes.
Real-world use case: Sales call preparation. Before every prospect call, run /research-prospect and /call-prep. Walk into conversations armed with relevant context, personalized talking points, and intelligent questions. Close rates improve because you sound like you actually understand their business.
Limitations: Quality depends entirely on how much information you feed it about your sales process and customers. Without proper configuration and historical data, outputs are generic and not particularly useful.
Pro tip: After each sales call, spend five minutes summarizing the conversation and feeding it back to Claude. This creates a knowledge base that makes future /call-prep outputs increasingly personalized and relevant. “Update CRM with notes from today’s call with Acme. They mentioned concerns about integration complexity and implementation timeline.”
5. Legal Plugin
What it does: Contract review, risk analysis, legal research, and document drafting.
Key commands:
/review-contract [file]– Analyzes contracts and flags issues/draft-nda [type]– Creates NDA templates/check-compliance [regulation]– Reviews documents for regulatory compliance
Configuration: Load your company’s standard contract language, legal playbooks (preferred clauses, deal breakers, acceptable risk levels), and relevant regulations for your industry.
This plugin scared investors for good reason—it can genuinely automate significant portions of legal review work. I tested /review-contract on a vendor MSA (Master Service Agreement). Claude:
- Highlighted standard clauses in green (no issues)
- Flagged concerning clauses in yellow with explanations
- Marked critical problems in red with specific modification recommendations
- Provided negotiation talking points for the yellow and red items
A junior associate would have taken 45 minutes for this analysis. Claude took 90 seconds. The quality was comparable—I had an actual attorney review the same contract, and Claude caught all the same issues plus two the attorney initially missed.
Real-world use case: First-pass contract review. Before sending contracts to legal counsel, use /review-contract to identify obvious issues. Attorneys then focus on complex, nuanced problems rather than catching basic liability clauses or missing indemnification provisions.
Limitations: This is NOT a replacement for actual legal counsel, especially for high-stakes agreements. Claude misses subtle issues, doesn’t understand industry-specific legal nuances, and can’t provide actual legal advice. It’s a triage tool, not a decision-maker.
Pro tip: Create multiple contract review profiles for different situations. One for vendor contracts (prioritizing data security and IP protection), one for customer contracts (balanced risk), one for partnership agreements (focusing on governance and exit clauses). Each profile teaches Claude different risk tolerances and priorities.
For deployment considerations and broader business implementation strategies, see our business implementation guide.
6. Finance Plugin
What it does: Financial analysis, modeling, reconciliation, and reporting.
Key commands:
/analyze-financials [file]– Reviews financial statements and highlights key metrics/build-model [scenario]– Creates financial projections/reconcile [accounts]– Matches transactions and identifies discrepancies
Configuration: Connect to accounting software (QuickBooks, Xero), define your key metrics and benchmarks, provide historical financial data for comparison.
I tested this with actual P&L statements from a client. The /analyze-financials output was sophisticated:
- Calculated margin trends month-over-month
- Flagged unusual expense categories (marketing spend spiked 40% in October)
- Compared metrics to industry benchmarks
- Identified cash flow concerns based on payment terms and collections
- Suggested areas for potential cost optimization
An experienced financial analyst would produce better insights, but this was solid work from an automated tool.
Real-world use case: Monthly financial review. Run /analyze-financials on monthly statements to catch anomalies and trends. The plugin doesn’t replace your CFO but gives them a head start on analysis, freeing time for strategic financial planning.
Limitations: Doesn’t understand business context that affects financial numbers. If revenue dropped because you intentionally paused a product line, Claude flags it as concerning without knowing the strategic rationale. Needs human context to interpret numbers correctly.
Pro tip: Build a financial narrative document explaining key decisions, strategic changes, and expected seasonal variations. Claude references this when analyzing financials, providing more contextually appropriate insights.
7. Data Analyst Plugin
What it does: Data querying, analysis, visualization, and reporting.
Key commands:
/query [database] [question]– Runs SQL queries to answer business questions/analyze [dataset]– Performs statistical analysis/visualize [data] [chart type]– Creates graphs and charts
Configuration: Connect to data warehouses (Snowflake, BigQuery, Redshift), define key datasets and what they contain, establish data governance rules (what can/cannot be queried).
This plugin is incredibly powerful for non-technical users who need data insights. I gave it access to a customer database and asked: “What are the characteristics of customers who churn versus those who renew?”
Claude:
- Wrote and executed SQL queries to segment customers
- Ran statistical analyses comparing cohorts
- Created visualizations showing key differences
- Identified that churned customers had three common characteristics: lower engagement in first 30 days, fewer users added to their accounts, and longer time-to-first-value
This type of analysis typically requires a data analyst spending a full day. Claude delivered actionable insights in 15 minutes.
Real-world use case: Self-service analytics for business teams. Marketing, sales, and product teams can answer their own data questions without creating tickets for the data team. “/query customer_db What were our top 10 performing marketing channels last quarter by CAC?”
Limitations: Requires proper data warehouse setup and governance. If your data is messy, poorly documented, or scattered across systems, Claude will struggle. Also, it sometimes writes inefficient queries that work but run slowly on large datasets.
Pro tip: Create a data dictionary explaining what each table and column means in business terms. Claude references this when generating queries, making results more accurate and meaningful.
8. Documentation Plugin
What it does: Creates technical documentation, API references, user guides, and knowledge base articles.
Key commands:
/document [codebase]– Generates technical documentation/create-guide [topic]– Writes user-facing guides/update-changelog [version]– Produces release notes
Configuration: Provide access to codebases (GitHub), existing documentation for style reference, and templates for different documentation types.
I tested this by asking it to document an internal API we’d built but never properly documented. Claude:
- Scanned the codebase and identified all endpoints
- Generated API reference documentation with parameters and examples
- Created code samples in multiple languages
- Wrote a quickstart guide for new developers
The quality was better than I expected—not perfect, but a solid draft that would have taken a technical writer a full week to produce from scratch.
Real-world use case: Keeping documentation current. When code changes, use /update-changelog to document what changed and /document [new feature] to generate initial docs. Technical writers refine rather than create from scratch.
Limitations: Struggles with context that isn’t in code or existing docs. If important design decisions or architectural rationale aren’t documented anywhere, Claude can’t infer them. Also, writing style tends toward dry and technical—usually needs editing for readability.
Pro tip: Maintain architecture decision records (ADRs) that explain why key technical choices were made. Claude incorporates these when generating documentation, producing richer, more contextual docs.
9. Customer Support Plugin
What it does: Drafts support responses, routes tickets, analyzes support trends.
Key commands:
/draft-response [ticket]– Creates reply following support playbook/route [ticket]– Determines correct team/person for the issue/analyze-trends– Identifies patterns in support requests
Configuration: Connect to support platform (Zendesk, Intercom, Help Scout), load knowledge base articles, define support tiers and routing rules, provide response templates and tone guidelines.
The /draft-response command is transformative for support teams. I fed it various customer questions:
Simple question about account settings → Claude drafted a clear, helpful response with step-by-step instructions Frustrated complaint about a bug → Claude acknowledged frustration, explained the issue, provided a workaround, and noted when a fix would ship Vague question that needed clarification → Claude asked appropriate follow-up questions
Response quality was consistently good—professional tone, accurate information, appropriate empathy. Not quite as good as your best support agent, but significantly better than your worst.
Real-world use case: Tier-1 support automation. For routine questions (password resets, feature explanations, billing questions), use /draft-response to generate replies. Support agents review and send rather than drafting from scratch. Handles 2X more tickets in the same time.
Limitations: Can’t handle truly complex technical issues or situations requiring real judgment. If a customer is threatening legal action or the situation involves nuanced policy interpretation, Claude’s responses are too simplistic.
Pro tip: Track which ticket categories get the best Claude-generated responses. Automate those completely while routing complex issues directly to humans. Over time, expand automation as you tune Claude’s capabilities for your specific support challenges.
10. Research Plugin
What it does: Hybrid search across sources, synthesis with citations, formatted report generation.
Key commands:
/research [topic]– Comprehensive research with source citations/deep-dive [question]– Exhaustive analysis with multiple perspectives/competitive-intel [company/product]– Competitive analysis and benchmarking
Configuration: Define trusted sources (specific websites, databases, internal repositories), set citation standards, establish fact-checking requirements.
I tested /competitive-intel on three of our competitors. Claude:
- Scraped their websites for product information
- Found recent news articles and press releases
- Analyzed their pricing and positioning
- Identified their strengths and weaknesses based on customer reviews
- Produced a 15-page competitive analysis document with proper citations
This replaced what would have been two full days of manual research.
Real-world use case: Market research and competitive intelligence. Before product launches, strategic planning, or major decisions, use /research to compile relevant information. Analysts focus on insight generation rather than information gathering.
Limitations: Only as good as its sources. If important information isn’t publicly available or in connected systems, Claude can’t find it. Also, sometimes cites sources that don’t fully support the claims made—always verify critical information.
Pro tip: Combine with the Enterprise Search plugin for hybrid internal/external research. “/research [topic] including our internal discussions and external market data” produces insights grounded in both your company’s knowledge and broader market intelligence.
11. Legal Risk Analysis Plugin
What it does: Specialized contract review focusing on regulatory compliance and enterprise risk.
Key commands:
/risk-assessment [document]– Identifies legal and compliance risks/regulatory-check [regulation]– Verifies compliance with specific laws/flag-liabilities– Highlights liability exposure
Configuration: Define your risk tolerance, load relevant regulations (GDPR, HIPAA, SOC2, etc.), provide examples of acceptable vs. unacceptable contract terms.
This is like a more specialized version of the general Legal plugin. I tested it on a data processing agreement. Claude:
- Flagged three GDPR compliance issues
- Identified unlimited liability clauses that exceeded our risk tolerance
- Noted missing data breach notification provisions
- Recommended specific language changes to address each issue
For companies in regulated industries, this plugin is invaluable for first-pass compliance review.
Real-world use case: Vendor contract review for regulated companies. Before legal counsel reviews contracts, use /regulatory-check GDPR (or HIPAA, SOX, etc.) to catch obvious compliance issues. Legal team focuses on complex interpretation rather than checkbox compliance.
Limitations: Regulations change frequently, and Claude’s knowledge might lag. For the most current regulatory requirements, you need to update the plugin’s knowledge base or verify its recommendations against current law.
Pro tip: Create regulation-specific review profiles. One for GDPR, one for HIPAA, one for SOC2, etc. Each contains the specific requirements and common pitfalls for that regulation. Switch profiles based on the contract type.
Building Custom Plugins
The official plugins cover many common use cases, but you’ll likely need customization for your specific workflows. Building custom plugins is surprisingly accessible, even for non-developers.
Plugin Structure:
Every plugin is a JSON file containing:
- Metadata (name, description, version)
- Skills (domain knowledge as text)
- Commands (shortcuts with underlying prompts)
- MCP connectors (API integrations)
- Sub-agent definitions (specialized workers)
You can download the official plugins from Anthropic’s GitHub and use them as templates.
Creating a Simple Custom Plugin:
Let’s walk through creating a basic custom plugin for content editing. We want commands that check for common writing issues.
{
"name": "Content Editor",
"version": "1.0",
"description": "Editorial review and content improvement",
"skills": [
"Professional editing standards",
"AP Style guidelines",
"SEO best practices",
"Readability optimization"
],
"commands": {
"/edit-for-clarity": {
"description": "Review text for clarity and readability",
"prompt": "Review the following text for clarity. Identify confusing sentences, jargon that needs explanation, and opportunities to simplify without losing meaning. Provide specific rewrite suggestions."
},
"/check-seo": {
"description": "Analyze content for SEO optimization",
"prompt": "Analyze this content for SEO. Check keyword usage, heading structure, meta description quality, and readability. Suggest specific improvements."
}
}
}
Save this as content-editor.json and import it into Cowork via /import-plugin [filename].
Advanced: Adding MCP Connectors:
For plugins that need external data, you’ll define MCP connectors. Here’s a simple example connecting to a company’s style guide stored in Notion:
"mcp_connectors": {
"style_guide": {
"type": "notion",
"database_id": "your-notion-database-id",
"query_fields": ["rule", "example", "category"],
"auth": "notion_oauth"
}
}
When commands reference the style guide, Claude queries this Notion database automatically.
Testing Custom Plugins:
Create a test workspace to try plugins before deploying to production:
- Load the plugin in a test workspace
- Run each command with sample data
- Verify outputs match expectations
- Check for edge cases and error handling
- Refine prompts based on actual results
Sharing Custom Plugins:
Currently, plugins save locally. To share with team members:
- Export the JSON file
- Document configuration requirements (API keys, data sources)
- Provide example usage for each command
- Establish versioning system for updates
Anthropic is building organization-wide plugin sharing, but it’s not available yet as of February 2026.
Plugin Best Practices and Advanced Tips
After weeks of intensive plugin usage, here are the patterns that separate effective from frustrating deployments.
Start with Official Plugins, Customize Later:
Don’t build custom plugins immediately. Use official ones first to understand what works and what doesn’t for your workflows. After you’ve internalized how plugins operate, customize or build from scratch.
One Plugin Per Major Function:
Resist the temptation to create a mega-plugin that does everything. Focused plugins are easier to maintain, debug, and improve. If you need multiple capabilities, use multiple specialized plugins rather than one overstuffed one.
Document Everything:
For each plugin, maintain documentation on:
- What it does and why it exists
- Configuration requirements
- Example commands and expected outputs
- Known limitations and workarounds
- Version history and changelog
This matters tremendously when team members need to use plugins you’ve created or when you revisit a plugin months later.
Iterate Based on Actual Usage:
Initial plugin configurations are never perfect. Track which commands get used frequently and which rarely. Refine heavily-used commands for better outputs. Consider removing rarely-used commands to reduce complexity.
Version Control Your Plugins:
Store plugin JSON files in Git or similar version control. This lets you:
- Track changes over time
- Roll back if updates break functionality
- Maintain different versions for different use cases
- Collaborate on plugin development
Test Edge Cases:
Don’t just test plugins with ideal inputs. Try:
- Incomplete or ambiguous data
- Exceptionally large datasets
- Edge cases in your domain
- Malformed input
- Contradictory instructions
Robust plugins handle edge cases gracefully rather than producing nonsensical outputs.
Conclusion: Plugins Are Where Cowork’s Power Lives
Base Claude Cowork is impressive. But plugins are what transform it from impressive to indispensable for specialized work. The difference between generic AI assistance and having a tool that actually understands your specific job comes down to plugin configuration.
The eleven official plugins provide excellent starting points for common business functions. Most companies will find 3-5 of these immediately useful with minimal customization. Another 2-3 become valuable after configuration. The rest might not match your needs at all—and that’s fine. You don’t need every plugin.
Custom plugin development is the next frontier. As teams gain experience with official plugins, they’ll identify gaps and build specialized tools for their unique workflows. The technical barrier is lower than you might expect—basic custom plugins require no coding, just JSON configuration and clear thinking about what Claude needs to know.
The ecosystem will mature rapidly. Expect third-party plugin marketplaces, industry-specific plugin collections, and dramatically expanded capabilities as the community shares innovations. Being early in this space means you’re building expertise that compounds as the platform evolves.
If you haven’t tested plugins yet, start this week. Install the Productivity plugin and use it for a few days. Once comfortable, add one domain-specific plugin (Legal, Finance, Marketing, whatever matches your work). Build familiarity gradually rather than trying to master everything at once.
The companies that win with Cowork won’t be those with the most plugins—they’ll be those who thoughtfully select, configure, and refine plugins that genuinely improve their specific workflows. Focus on quality over quantity, actual usage over theoretical capability, and continuous refinement over set-it-and-forget-it deployment.