AI Assistant Connector (MCP Server) - Introduction

CanIPhish can now be driven directly from an AI assistant such as Claude or ChatGPT using the Model Context Protocol (MCP). Once connected, you can run your entire security-awareness program in plain English — “show me this quarter’s phishing results”, “which employees are highest risk?”, “create and schedule quarterly training for all staff” — and the assistant carries it out in CanIPhish, with your approval.

This article explains what the MCP connector is, what it can do, how to connect it, and how it keeps your program safe.

What is the CanIPhish MCP connector?

The Model Context Protocol is an open standard that lets AI assistants securely connect to external tools and data. The CanIPhish connector exposes your tenant to your assistant as a set of tools it can use on your behalf.

In practice, this means you can ask your AI assistant to read your program and take action in CanIPhish — creating campaigns, building content and employee lists, pulling reports, triaging reported emails, and more — without clicking through the platform yourself. The connector:

• Authenticates as you. The assistant acts with your permissions, scoped to the tenant you choose when connecting.
• Only acts with your approval. You approve the connection on a consent screen, and your assistant asks before it makes any change or sends anything.
• Works with any MCP-capable client — Claude, ChatGPT, VS Code, Cursor, and others.

What you can do with it

Once connected, you can ask your assistant to work across your whole program. A few examples of what’s possible:

The assistant discovers the right tool for each request automatically — you just describe what you want.

Enterprise capabilities. On Enterprise plans, the assistant can also use Auto-Phish and Auto-Train, where CanIPhish’s AI Program Manager selects the most contextually appropriate phishing lure and training module for each individual employee based on their role, history and risk — so you can launch an adaptive program without hand-picking content.

Before you begin

You’ll need:

1. A Professional or Enterprise CanIPhish plan. The AI Assistant Connector is available on these plans. (On other plans you’ll see an upgrade prompt where the connector settings would appear.)
2. An administrator account. Only Account Admins and Super Admins can connect an assistant.
3. An AI client that supports MCP — for example the Claude or ChatGPT desktop/web apps, VS Code, or Cursor.

How to connect

Step 1 — Copy your MCP URL

In CanIPhish, open your User Profile (available at /platform/settings) and select the AI Assistant Connector (MCP) menu item. Copy the MCP URL:

https://api.caniphish.com/api/v3/mcp

Step 2 — Add the connector in your AI client

Use the steps for your client:

• Claude — Settings → Connectors → Add custom connector → paste the MCP URL → Connect.
• ChatGPT
  • Step 1. Settings → Plugins → Developer mode → Enable Developer mode.
  • Step 2. Settings → Plugins → Browse plugins → Click '+' symbol
  • Step 3. Configure the following settings and click Create:
    • Icon: https://assets.caniphish.com/assets/images/icon-small.png (Download and Upload)
    • Name: CanIPhish
    • Description: CanIPhish MCP Connector
    • Connection: https://api.caniphish.com/api/v3/mcp
    • Authentication: OAuth (Default)
• Cursor — use the “Add to Cursor” link shown in the connector settings to add it automatically.

Step 3 — Sign in and choose your tenant

When the client connects, you’ll be taken to a CanIPhish consent screen. Here you:

1. Sign in (or paste your CanIPhish API key, shown in the same settings area), and
2. Choose which tenant the assistant is allowed to access.

Approve the connection, and your assistant is ready to use. That’s it — you can now start asking it to work in CanIPhish.

How it keeps your program safe

Because these tools act on real employees, the connector is built around explicit, layered control:

• You approve the connection. Nothing is exposed until you sign in and consent, and the assistant is limited to the tenant you selected.
• It acts as you. The assistant inherits your role and permissions — it can’t do anything you couldn’t do yourself.
• Reading is safe; changing asks first. Read-only requests (reports, overviews, risk data) run smoothly, but any tool that creates, changes, or sends something prompts your assistant to confirm with you first.
• Sending real phishing needs a second confirmation. Before an assistant can send a live phishing campaign, it must first preview exactly who will be targeted and obtain a one-time confirmation — so a send can never happen without you seeing the recipient list.
• Revoke any time. You can disconnect from your AI provider, or from the Connected assistants list in the connector settings.

Working with multiple tenants

If your CanIPhish access spans more than one tenant (for example, an MSP or partner managing several client tenants), the assistant will ask you which tenant to work in before it runs anything, and it will target that tenant explicitly on every action. This prevents a request meant for one client from ever affecting another. Just tell it which tenant you mean — e.g. “work in the Acme Corp tenant.”

Managing and revoking connections

• To see what’s connected, open the AI Assistant Connector (MCP) settings and review the Connected assistants list.
• To revoke access, remove the connection there, or remove the CanIPhish connector from within your AI provider’s settings. Revoked connections lose access immediately.

Example prompts to get started

• “Give me a snapshot of our security-awareness program — phishing results, training completion, and the leaderboard.”
• “List our five highest-risk employees and why.”
• “Create an immediate phishing test for the Finance team using a payroll-themed lure. Send me a test email before it goes live.”
• “Set up recurring monthly training for all staff, due within 14 days.”
• “What onboarding steps haven’t we completed, and what should we prioritise next?”
• “Triage today’s reported emails and flag anything that looks like a genuine threat.”

Troubleshooting

• The connector settings aren’t showing. The AI Assistant Connector is only visible to Account Admins / Super Admins on Professional and Enterprise plans. Check your plan and role.
• The connection failed during sign-in. Reopen the connector in your AI client and try again, making sure you paste the full MCP URL exactly as shown above. If you use a paste-key sign-in, confirm your API key is current.
• The assistant says it needs a tenant. Your access spans multiple tenants — tell it which one to use (e.g. “use the Acme Corp tenant”).
• A phishing send didn’t go out. Live sends require the preview-and-confirm step. Ask the assistant to propose the campaign, review the recipient list, and confirm.

Frequently asked questions

Does the assistant have access to everything in my account?
It has the same access you do, limited to the single tenant you approve at connection time.

Can it send phishing emails or assign training without me knowing?
No. Changes require your assistant to confirm with you, and live phishing sends require a separate preview-and-confirm of the exact recipient list.

Which assistants are supported?
Any client that supports the Model Context Protocol, including Claude, ChatGPT, VS Code and Cursor.

How do I turn it off?
Revoke the connection from the Connected assistants list in CanIPhish, or remove the connector in your AI provider — access ends immediately.

Need a hand getting connected? Contact CanIPhish support and we’ll walk you through it.