Skip to main content

Plugins and Connectors: Extending Cowork's Reach

In Lesson 28, you explored browser integration for web-based workflows. Now you will connect Cowork to the services where your real data lives (Google Drive, Slack, Jira, and dozens more) through Plugins and Connectors.

You learned about MCP (Model Context Protocol) in Lesson 12, where developers create servers that expose tools to Claude Code. Plugins and Connectors bring the same capability to Cowork, but without requiring any development work.


What Plugins Are

A Plugin is a workflow package that bundles multiple capabilities into a single, installable unit. Think of it like an app on your phone: one install gives you everything you need.

A Plugin can contain any combination of:

ComponentWhat It DoesExample
ConnectorsLink to external data sourcesGoogle Drive access, Slack messages
SkillsAdd specific capabilitiesData analysis, report generation
Slash commandsCreate shortcuts for common tasks/weekly-report, /standup-summary
Sub-agentsEnable autonomous workflowsResearch assistant, document reviewer

You install a Plugin from the + button > Plugins > Add plugin menu in the desktop app, or via the CLI. All its components become available in your session.

Department-specific Plugin templates provide ready-to-use packages for common roles:

  • HR: Employee onboarding checklists, policy lookup, benefits Q&A
  • Finance: Expense report analysis, budget tracking, compliance checks
  • Legal: Contract review, clause comparison, regulatory lookup
  • Design: Asset management, design system reference, feedback collection
  • Engineering: Code review summaries, incident response, documentation
  • Operations: Process automation, vendor management, reporting

These templates are starting points. You can customize them by adding or removing individual components.


What Connectors Are

Connectors are one component within a Plugin. They are pre-packaged remote MCP servers that link Claude to external data sources.

Here is how the layers relate:

  • Protocol layer: MCP (Model Context Protocol); the open standard
  • Integration layer: Connector; a single data source connection built on MCP
  • Workflow Package layer: Plugin; bundles Connectors + skills + slash commands + sub-agents

The Anthropic Connectors Directory lists 50+ Connectors across categories:

ConnectorData SourceWhat It Provides
Google DriveGoogle WorkspaceRead, search, and modify documents
Google CalendarGoogle WorkspaceView and manage calendar events
GmailGoogle WorkspaceRead and search email
NotionNotion workspaceAccess pages, databases, and docs
SlackSlack workspaceRead messages, search conversations
GitHubGitHub repositoriesRead code, issues, and discussions
JiraAtlassian JiraQuery tickets, update status
SalesforceCRM dataAccess accounts, opportunities, reports
DocuSignDocument signingManage agreements and signatures
FigmaDesign filesAccess designs, components, and comments
CanvaDesign platformBrowse and reference design assets
WordPressContent managementManage posts, pages, and media
ApolloSales intelligenceContact and account data enrichment
ClayData enrichmentAutomated lead research and scoring
OutreachSales engagementSequence management and analytics
SimilarwebWeb analyticsTraffic and competitive intelligence
MSCIFinancial dataESG ratings and market analytics
FactSetFinancial researchMarket data and company analysis
LegalZoomLegal servicesDocument templates and filings
HarveyLegal AILegal research and analysis

You don't write code. You don't configure servers. You authenticate, grant permissions, and Claude can access the data.

Connectors use remote MCP servers; hosted in the cloud by Anthropic and partners; rather than local servers running on your machine. This means Connectors work across all Claude platforms: web, desktop, mobile apps, and API.


MCP vs. Connectors: What's the Difference?

MCP (Lesson 12) is for developers building custom integrations:

  • Requires programming (Python, JavaScript, etc.)
  • You design the tools and data structures
  • You host and maintain the MCP server
  • Full control over the integration
  • Best for: proprietary data sources, custom APIs

Connectors are for knowledge workers using common services:

  • No programming required
  • Pre-defined tools and data structures
  • Anthropic and partners handle maintenance via remote MCP servers
  • Optimized for popular services
  • Best for: widely-used SaaS platforms

The relationship: Connectors are MCP servers. Someone else built them, packaged them, and maintains them. You just use them.


How Connectors Work

When you add a Connector to Cowork:

  1. Authentication: You sign in to the external service (Google, Notion, etc.)
  2. Permission Grant: You authorize what Claude can access
  3. Tool Registration: The Connector exposes its capabilities as tools
  4. Querying: Claude can now query, read, and sometimes modify data

From that point forward, Claude can reference data from the connected service alongside your local files.

Example: With the Google Drive Connector, you could ask:

"Look at the project planning document in my Google Drive, compare it to the local project files I showed you, and tell me what's missing from the local version."

Claude reads the Google Doc via Connector, reads your local files, and performs the comparison; all without you manually copying anything.

Some Connectors also render interactive apps inline in the chat. For example, a calendar Connector might display an interactive calendar widget where you can select dates directly, rather than typing them out.


Setting Up Connectors

Step 1: Open Connector Settings

In Claude Desktop:

  1. Click the + button next to the prompt box
  2. Select Connectors from the menu
  3. You'll see available Connectors organized by category

You can add connectors before or during a session. To manage or disconnect connectors later, go to Settings > Connectors in the desktop app, or select Manage connectors from the Connectors menu in the prompt box.

Step 2: Add a Connector

  1. Click "Add" next to the service you want to connect
  2. A browser window opens for authentication
  3. Sign in and authorize Claude's access
  4. Return to Claude Desktop; connection confirmed

Step 3: Configure Permissions

Each Connector has permission scopes:

  • Read-only: Claude can view data but not modify
  • Read-write: Claude can modify data (use with caution)
  • Specific resources: Limit access to specific folders or workspaces

Start with read-only access. Only enable read-write when you trust the workflow and understand what Claude will do.

Access Tiers

Plugins and Connectors require a paid plan:

  • Pro, Max, Team, Enterprise: Full Connector installation, custom Connectors via MCP URL, and interactive apps
  • Free plan: No access to Plugins or Connectors

The Combination Power

Connectors shine when combined with local file operations:

Scenario: You're preparing a quarterly report. The data lives in:

  • Google Sheets (sales figures)
  • Notion (product updates)
  • Slack (customer feedback)
  • Local files (previous quarter's report template)

Without Connectors: You download exports from each service, copy-paste into your document, and hope nothing changes.

With Connectors:

"Create a quarterly report using the template in my local files. Pull sales figures from the Q4 Sales Google Sheet, include product updates from the Notion product database, summarize customer feedback from the #customers Slack channel, and compare everything to last quarter's performance."

Claude:

  1. Reads the local report template
  2. Queries Google Sheets for current sales data
  3. Fetches Notion pages for product updates
  4. Searches Slack for customer feedback
  5. Analyzes everything and generates the report

The advantage: Live data, no manual export/import, and one request does the work of accessing four different systems.


Available Connectors

Document and Knowledge:

  • Google Drive (Docs, Sheets, Slides)
  • Notion
  • Confluence
  • SharePoint

Communication:

  • Slack
  • Microsoft Teams
  • Gmail
  • Google Calendar

Development:

  • GitHub
  • GitLab
  • Linear

Business Data:

  • Salesforce
  • HubSpot
  • Jira
  • Airtable
  • DocuSign

Design:

  • Figma
  • Canva

Content:

  • WordPress

New Connectors are added regularly. You can browse available integrations from the + button > Connectors menu inside Claude Desktop, with 50+ and growing.


Slash Commands and Structured Forms

Plugins can expose slash commands: shortcuts you type in the chat input (like /weekly-report or /standup-summary). When you invoke a slash command, Cowork can present a structured form (a fill-in UI with labeled fields) instead of requiring you to type a free-text prompt. You fill in the form fields, click submit, and the Plugin executes the workflow with your inputs.

This matters because structured forms reduce ambiguity. Instead of hoping Claude interprets your free-text prompt correctly, the form guides you to provide exactly the inputs the workflow needs.


Building Custom Plugins

You don't have to wait for Anthropic or partners to build a Plugin for your workflow. The Plugin Create tool lets you build custom Plugins directly inside Cowork:

  1. Describe what your Plugin should do
  2. Claude generates the Plugin structure (Connectors, skills, slash commands)
  3. You test, refine, and publish

Private GitHub repositories can also serve as Plugin sources (currently in beta). This means your team can maintain internal Plugins in version control, with updates deployed through standard git workflows.


Enterprise Features

For organizations deploying Plugins at scale:

  • Organization marketplace: Admins curate and distribute approved Plugins across the company
  • OpenTelemetry tracking: Monitor Plugin usage, performance, and errors through standard observability tools
  • Per-user provisioning: Assign specific Plugin sets to roles or departments automatically
  • Auto-install policies: New team members get the right Plugins on day one without manual setup

These features mean Plugins scale from individual productivity to organization-wide automation.


Portability: Plugins Beyond Cowork

A key architectural insight: Plugins are portable file systems you own. They aren't locked to the Cowork desktop interface. The same Plugin works across:

  • Cowork (desktop workflows)
  • Claude Agent SDK (programmatic agent orchestration)
  • Any future Claude surface that supports the Plugin format

This portability means investing in Plugin creation isn't a bet on a single interface. The workflows you build today carry forward as Claude's platform evolves.


Current Limitations

Plugins and Connectors are powerful but have constraints:

Rate Limits: External APIs have usage limits. Claude queries efficiently, but massive data pulls may hit limits.

Authentication: Some services require re-authentication periodically. You'll be prompted when this happens.

Read-only vs. Read-write: Not all Connectors support modification. Check capabilities before planning write workflows.

Service availability: If the external service is down, the Connector won't work.

Data freshness: Connectors fetch current data, not real-time streams. Changes after Claude queries won't be reflected.


When to Use Plugins and Connectors

Ideal for:

  • Reports combining data from multiple sources
  • Research that spans across platforms
  • Cross-reference analysis (e.g., GitHub issues vs. Jira tickets)
  • Automated reporting from SaaS platforms
  • Department-specific workflows using Plugin templates

Less ideal for:

  • Real-time monitoring (use dedicated dashboards)
  • Massive data exports (use native export features)
  • Complex data transformations (export, process locally)

Privacy and Security

Plugins and Connectors require granting Claude access to your external accounts. Consider:

Principle of Least Privilege: Grant only the access needed. Read-only for reporting, specific folders rather than entire workspaces.

Regular Audits: Periodically review which Connectors are active and revoke access you no longer need.

Sensitive Data: Be cautious connecting accounts with highly sensitive information (HR data, financial systems).

Service Terms: Ensure using Plugins and Connectors complies with your organization's policies on external tool access.


Try With AI

Map Your Data Sources to Connectors:

"List the 5 services where I spend the most time accessing data for my work (e.g., Google Drive, Slack, Jira, Salesforce). For each one, check whether a Connector exists in the Anthropic directory (50+ available). Then identify which 2-3 would save the most time if Claude could query them directly in a single prompt; and describe one specific multi-source task I do today that would benefit."

What you're learning: Data source mapping; connecting your real workflow to the Connector ecosystem. This is the first step to replacing manual export/import with live multi-source queries.

Design a Plugin for Your Role:

"Based on what you know about my work, design a custom Plugin for my role. Include: (1) which Connectors it would bundle, (2) two slash commands with structured form fields for tasks I repeat weekly, (3) one custom skill for domain reasoning I apply often. Explain why this bundle is more valuable than using each component separately."

What you're learning: Plugin architecture thinking; understanding how bundling Connectors, skills, and slash commands into a single package creates compound value. This is the mental model that separates using Cowork from designing workflows in Cowork.


What's Next

Plugins and Connectors extend Cowork's reach to external data sources and bundled capabilities. But Cowork is still evolving. The next lesson covers current limitations, safety considerations, and what's coming in future updates, including Knowledge Bases that will give Claude persistent memory across sessions.

Flashcards Study Aid