AI & Modern CRM6 min read

How to Connect Claude / ChatGPT to Your CRM via MCP

A step-by-step guide to setting up the Model Context Protocol so your AI assistant can search contacts, create deals, log activities, and query your pipeline — all through natural conversation.

Key Takeaways

MCP (Model Context Protocol) is the bridge that lets AI assistants talk to your CRM using structured tools.
Setup requires three things: an MCP-compatible CRM, an AI client that supports MCP, and an API key.
Once connected, you can manage contacts, deals, and activities through natural conversation.
Security is enforced by your CRM's API key scoping — the AI can only do what the key permits.

What You Need

Connecting an AI assistant to your CRM via MCP requires three components:

An MCP-compatible CRM

Your CRM must expose an MCP server — a structured interface that defines what tools (actions) the AI can use. Not all CRMs support this yet.

An MCP client

The AI application that connects to the MCP server. Claude Desktop, Cursor, and Windsurf all support MCP natively. Custom agents can use the MCP SDK.

An API key

Authentication credential that lets the MCP server verify your identity and scope what the AI can access. Treat this like any other API key — keep it secure.

Step-by-Step Setup

Step 1: Generate an API key

In your CRM settings, generate an API key specifically for MCP use. Best practice: create a dedicated key for AI access (separate from your integration keys) so you can revoke or audit it independently.

💡Security tip

Use the principle of least privilege. If your CRM supports scoped keys, give the MCP key access only to the operations the AI needs. A key used for read-only pipeline queries doesn’t need write access to contacts.

Step 2: Configure your MCP client

Most MCP clients use a JSON configuration file to define which servers to connect to. The typical configuration looks like this:

{
  "mcpServers": {
    "your-crm": {
      "command": "npx",
      "args": ["-y", "@your-crm/mcp-server"],
      "env": {
        "CRM_API_KEY": "your-api-key-here"
      }
    }
  }
}

The command and args will vary by CRM. Your CRM’s MCP documentation will provide the exact values.

Step 3: Verify the connection

Once configured, open your AI client and try a simple query like “list my recent contacts” or “show my open deals.” The AI should use the CRM’s MCP tools to fetch and display real data.

What You Can Do Once Connected

Natural Language	What the AI Does
"Find all contacts at Acme"	Calls search_contacts with company filter
"Create a deal for the Acme renewal, $50K"	Calls create_deal with title, value, and company
"Log a call with Jane — discussed Q3 timeline"	Calls log_activity with contact, type, and notes
"What's our pipeline look like for Q3?"	Calls get_forecast or search_deals with date filter
"Move the Acme deal to negotiation stage"	Calls update_deal with new stage
"Tag everyone from the conference as 'SaaStr 2026'"	Calls bulk_tag with contact list and tag name

⚠️Trust but verify

When using AI to write data (create contacts, move deal stages), always review the AI’s proposed action before confirming. Most MCP clients show you what the AI plans to do and ask for confirmation before executing write operations.

How crmzero handles this

crmzero’s MCP server is available as an npm package. Connect in under 2 minutes:

{
  "mcpServers": {
    "crmzero": {
      "command": "npx",
      "args": ["-y", "crmzero-mcp"],
      "env": {
        "CRMZERO_API_KEY": "your-key"
      }
    }
  }
}

58 tools covering contacts, deals, activities, tags, forecast, and pipeline operations. Works with Claude Desktop, Cursor, and any MCP-compatible client.

See full MCP setup guide →

Frequently Asked Questions

Can I use MCP with ChatGPT?

ChatGPT doesn't natively support MCP yet, but you can use MCP with Claude Desktop, Cursor, Windsurf, and custom agents built with the MCP SDK. As MCP adoption grows, more clients will add support.

Is it safe to give AI write access to my CRM?

Yes, with proper precautions. Use scoped API keys, limit the AI to necessary operations, and use an MCP client that shows confirmations before executing write operations. Most enterprise CRMs also maintain audit logs of all API operations.

What happens if the AI makes a mistake?

MCP tools are deterministic — the AI calls specific functions with specific parameters. If the AI misinterprets your request, the action can usually be reversed (e.g., undo a stage change, update a contact). This is different from AI 'hallucinating' because the CRM executes exact operations, not free-form text.

Do I need coding skills to set up MCP?

Basic setup requires editing a JSON configuration file — no coding needed. If you want to build custom MCP workflows or run an MCP server programmatically, familiarity with Node.js helps.

How much does MCP cost?

MCP itself is a free, open protocol. The cost depends on your CRM plan (whether it offers MCP server access) and what AI client you use. Most CRMs gate MCP access on higher-tier plans.

AI & Modern CRM