Skip to main content
Connectors wire Claude to the external tools your role depends on — CRMs, project trackers, data warehouses, design tools, and more. They enable Claude to read data from and write data to your company’s systems.

​
How connectors work

Connectors use the Model Context Protocol (MCP) to provide standardized access to external tools. Each connector is an MCP server that exposes tools, resources, and prompts.
1

Configure in .mcp.json

List the MCP servers your plugin should connect to
2

Claude discovers capabilities

When the plugin loads, Claude queries each server to discover available tools
3

Skills and commands use tools

Your plugin’s skills and commands can invoke tools from connected servers
4

Data flows bidirectionally

Claude can read data (query CRM), write data (update deal stage), or both

​
Configuration

Connectors are defined in your plugin’s .mcp.json file:
.mcp.json
{
  "mcpServers": {
    "hubspot": {
      "type": "http",
      "url": "https://mcp.hubspot.com/anthropic"
    },
    "slack": {
      "type": "http",
      "url": "https://mcp.slack.com/mcp"
    },
    "notion": {
      "type": "http",
      "url": "https://mcp.notion.com/mcp"
    },
    "fireflies": {
      "type": "http",
      "url": "https://api.fireflies.ai/mcp"
    }
  }
}

​
Server types

​
type
string
required
The MCP server transport type. Currently supports http for HTTP-based servers.
​
url
string
required
The endpoint URL for the MCP server.

​
Real example: Sales plugin connectors

The sales plugin connects to tools across the sales stack:
sales/.mcp.json
{
  "mcpServers": {
    "hubspot": {
      "type": "http",
      "url": "https://mcp.hubspot.com/anthropic"
    },
    "close": {
      "type": "http",
      "url": "https://mcp.close.com/mcp"
    },
    "clay": {
      "type": "http",
      "url": "https://api.clay.com/v3/mcp"
    },
    "zoominfo": {
      "type": "http",
      "url": "https://mcp.zoominfo.com/mcp"
    },
    "fireflies": {
      "type": "http",
      "url": "https://api.fireflies.ai/mcp"
    },
    "slack": {
      "type": "http",
      "url": "https://mcp.slack.com/mcp"
    }
  }
}

​
What each connector provides

Tools:
  • Search accounts, contacts, deals, activities
  • Create and update records
  • Log calls, emails, meetings
  • Manage pipeline stages
Use cases:
  • Pull account history for call prep
  • Update deal stages after meetings
  • Create follow-up tasks
  • Log activity from call summaries
Tools:
  • Search messages across channels
  • Read channel history
  • Post messages
  • Upload files
Use cases:
  • Find internal discussions about accounts
  • Share pipeline reviews with team
  • Get colleague insights for call prep
Tools:
  • Search call recordings
  • Retrieve transcripts
  • Extract key moments
  • Get speaker analytics
Use cases:
  • Process call transcripts automatically
  • Review prior conversations before new calls
  • Extract competitive intelligence
Tools:
  • Company lookup
  • Contact enrichment
  • Find email addresses
  • Get firmographic data
Use cases:
  • Research prospects for outreach
  • Verify contact information
  • Build account context

​
Standalone vs supercharged

Plugins are designed to work with or without connectors:
Works without any connectorsClaude relies on:
  • Manual input from you
  • Web research
  • File uploads
  • Your direct knowledge
Example: Call prep
  • You tell Claude: company name, attendees, meeting type
  • Claude does web research: recent news, funding, LinkedIn
  • Output: prep brief with research and suggested questions

​
How skills use connectors

Skills reference connectors in their execution flow:
call-prep/SKILL.md
## Execution Flow

### Step 1: Gather Context

**If connectors available:**

1. Calendar → Find upcoming meeting matching company name
   - Pull: title, time, attendees, description, attachments

2. CRM → Query account
   - Pull: account details, all contacts, open opportunities
   - Pull: last 10 activities, any account notes

3. Email → Search recent threads
   - Query: emails with company domain (last 30 days)
   - Extract: key topics, open questions, commitments

4. Chat → Search internal discussions
   - Query: company name mentions (last 30 days)
   - Extract: colleague insights, competitive intel

**If no connectors:**

1. Ask user:
   - "What company are you meeting with?"
   - "What type of meeting is this?"
   - "Who's attending?"
Claude automatically detects which connectors are available and adjusts its approach.

​
Connector tables in skills

Skills document which connectors enhance them: 
## Connectors (Optional)

| Connector | What It Adds |
|-----------|-------------|
| **CRM** | Account details, contact history, open deals, recent activities |
| **Email** | Recent threads with the company, open questions, attachments shared |
| **Chat** | Internal chat discussions about the account, colleague insights |
| **Transcripts** | Prior call recordings, topics covered, competitor mentions |
| **Calendar** | Auto-find the meeting, pull attendees and description |

> **No connectors?** No problem. Just tell me about the meeting and paste 
> any context you have. I'll research the rest.

​
Available connector types

Common categories of connectors used in knowledge work plugins:

CRM & Sales

  • HubSpot
  • Salesforce
  • Close
  • Outreach
  • Apollo

Communication

  • Slack
  • Microsoft Teams
  • Gmail
  • Outlook

Project Management

  • Jira
  • Linear
  • Asana
  • Monday
  • ClickUp

Data & Analytics

  • Snowflake
  • Databricks
  • BigQuery
  • Amplitude
  • Pendo

Documentation

  • Notion
  • Guru
  • Confluence
  • Google Docs

Meetings & Calls

  • Fireflies
  • Gong
  • Chorus
  • Zoom

Design

  • Figma
  • Canva
  • BioRender

Enrichment

  • ZoomInfo
  • Clay
  • Clearbit
  • Apollo

​
Swapping connectors

Customize which tools your plugin connects to:
1

Edit .mcp.json

Open your plugin’s .mcp.json file
2

Add your tools

Replace the default connectors with your company’s stack:
{
  "mcpServers": {
    "salesforce": {
      "type": "http",
      "url": "https://mcp.salesforce.com/api"
    },
    "teams": {
      "type": "http",
      "url": "https://teams.mcp.claude.com/mcp"
    }
  }
}
3

Update skill references

If skills reference specific connectors by name, update those references to match your new tools
The connector must provide an MCP server. Check the tool’s documentation or the MCP directory for availability.

​
Authentication

Authentication is handled at the MCP server level, not in your plugin configuration. When you connect to an MCP server:
  1. First connection — Claude prompts you to authenticate with the service
  2. Credentials stored — Tokens are securely stored by Claude
  3. Automatic refresh — Claude handles token refresh automatically
  4. Scoped access — You control what permissions the connector has

​
Data privacy and security

Connectors give Claude access to company data. Review what data each connector can access before enabling it.
What connectors can do:
  • Read data you have access to in connected systems
  • Write data when you explicitly approve actions
  • Search across your connected tools
What connectors cannot do:
  • Access data you don’t have permissions for
  • Make changes without your confirmation (when configured)
  • Share data between different companies/workspaces
Best practices:
  1. Only connect tools your role actually needs
  2. Review the permissions each connector requests
  3. Use read-only connectors when possible
  4. Audit connector activity regularly

​
MCP protocol deep dive

For details on how the Model Context Protocol works:

MCP Integration

Learn about the Model Context Protocol, server discovery, tool invocation, and building custom MCP servers

​
Best practices

Skills and commands should work without connectors, just less powerfully. Never make connectors a hard requirement.
Use connector tables in skills to show what’s enhanced when tools are connected.
Customize .mcp.json to reflect what your team actually uses. Don’t keep default connectors you don’t have.
Some workflows only need to read data. Configure appropriate permissions for each connector.
Validate your plugin works without connectors before adding tool integrations. This ensures a good baseline experience.

​
Next steps

MCP integration

Deep dive on the Model Context Protocol

Build a connector

Create a custom MCP server

Customize plugins

Swap connectors for your tool stack

Available connectors

Browse all supported MCP servers

Build docs developers (and LLMs) love

Get started for freeTalk to us