Cannon VoIP Help

Connect an AI agent (Claude, Gemini, or OpenAI)

Let your own AI read your calls and texts and act on your account, safely and with your approval.

Updated September 16, 2026

You can connect your own AI assistant - Claude, Gemini, OpenAI, or any tool that speaks MCP (the Model Context Protocol) - to your phone system. Once connected it can do things like read your voicemails, list who called, and text an existing conversation, and, with your approval, take bigger actions like setting your company to Holiday.

The golden rule: an agent can only do what you allow, can never do more than the person who set it up, and everything it does is written to your audit log. High-risk actions never happen on the agent's word alone - a person approves each one.

Set it up

  1. In the portal, open AI Agents (under Company). You need the same permission as creating API keys.
  2. Choose Connect an agent. Give it a name (for example "Claude - Front Desk"), tick what it may access, and, if you want, narrow it to specific tools.
  3. Choose how far it may act on its own:
  4. - Read only - it can look, not touch. - Low-risk actions - it can reply in an existing text thread, add a company contact, add a holiday. - High-risk actions (with approval) - it can also propose things like changing company status or texting a new number, but each one waits for a person to approve it.

  5. Pick where approval requests go: a push notification, a Team Chat post, or an email.
  6. Create the connection. You are shown a token once - copy it now.

Point your AI at it

Your MCP address is your portal web address with /mcp on the end - for example https://app.cannonvoip.com/mcp. In your AI's connector or MCP settings, add a connection to that address and set its authorization header to Bearer followed by the token you copied. That is all it needs; the same address and token work for Claude, Gemini, and OpenAI.

Approving what an agent proposes

When an agent asks to do something high-risk, it shows up under Waiting on your approval on the AI Agents page (and as the notification you chose). Review the plain-English summary and choose Approve or Deny. Approving runs the action under your authority; denying stops it. A request that nobody acts on expires by itself after an hour.

Turning it off

Any connection can be disconnected instantly with the Disconnect button. The token stops working the moment you do.

Was this helpful?
← All help articles