Set up an AI agent

visiconvert mcp runs a Model Context Protocol server on stdio. It offers the same operations as the command line, so an agent can check your setup, register tracking numbers, review leads and record bookings. The tool reference lists every tool.

Before you start

Install the CLI and log in. The MCP server uses the same saved login. See The visiconvert command line.

Claude Code

claude mcp add visiconvert -- visiconvert mcp --allow-changes

Leave off --allow-changes to give the agent read-only access.

Claude Desktop and other clients

Add the server to the client’s MCP configuration. For Claude Desktop, that is claude_desktop_config.json:

{
  "mcpServers": {
    "visiconvert": {
      "command": "visiconvert",
      "args": ["mcp", "--allow-changes"]
    }
  }
}

What the agent can do

Tier	Tools	When
Read	get_status, list_numbers, list_twilio_numbers, list_leads, get_lead_outcomes, source_report	Always
Change	add_tracking_number, add_lead, qualify_lead, record_outcome	Only with --allow-changes

Every tool carries MCP annotations that mark it as read-only or not, so clients can ask before changes. Without --allow-changes, the change tools are not listed at all.

What it will not do

• Handle Twilio credentials. No tool accepts them. Connecting Twilio stays a step you run yourself with visiconvert connect twilio, so your Auth Token never enters the agent’s context.
• Buy or release numbers, or delete leads and bookings.

Guided setup

The server includes a prompt, set_up_call_tracking. In clients that support MCP prompts, it walks the agent through checking status, choosing which campaigns to track, assigning numbers and confirming a test call. It asks you before each change and tells you to connect Twilio yourself.

You can also just ask, for example:

Set up call tracking for Google Ads, Facebook and our yard signs. Calls should ring through to our main line.