AgentPhone MCP Server

Give AI agents real phone numbers, SMS, and voice calls via the Model Context Protocol.

AgentPhone lets your AI agent buy phone numbers, send/receive SMS, and place voice calls — all through natural language in Cursor, Claude Desktop, or any MCP-compatible client.

Quick Start

1. Get your API key

2. Add to Cursor or Claude Desktop

Both use the same config. Add this under mcpServers:

Cursor: Settings > MCP or ~/.cursor/mcp.json
Claude Desktop: ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows)

{
  "mcpServers": {
    "agentphone": {
      "command": "npx",
      "args": ["-y", "agentphone-mcp"],
      "env": {
        "AGENTPHONE_API_KEY": "your_api_key_here",
        "AGENTPHONE_BASE_URL": "https://api.agentphone.to"
      }
    }
  }
}

You can also run the built server with node and a path to dist/index.js (e.g. after cloning and building this repo).

What Can It Do?

Once configured, just ask your AI agent things like:

"Buy me a US phone number"
"Show me my recent calls and transcripts"
"Call +14155551234 and have a conversation about scheduling a dentist appointment"
"Create an agent called Support Bot and give it a number"
"Show me the latest SMS messages"
"Set up a webhook to receive inbound messages"

Available Tools

Phone Numbers

| Tool | Description | |------|-------------| | list_numbers | List all phone numbers in your account | | buy_number | Purchase a new phone number | | release_number | Release a phone number (irreversible) |

SMS

| Tool | Description | |------|-------------| | get_messages | Get SMS messages for a number | | list_conversations | List SMS conversations | | get_conversation | Get a conversation with full message history |

Voice Calls

| Tool | Description | |------|-------------| | list_calls | List recent calls (inbound and outbound) | | get_call | Get call details and transcript | | make_call | Place an outbound call (uses your webhook for responses) | | make_conversation_call | Place a call with a built-in AI conversation on a topic you define |

Agents

| Tool | Description | |------|-------------| | list_agents | List all agents | | create_agent | Create a new agent | | get_agent | Get agent details and attached numbers | | attach_number | Assign a phone number to an agent |

Webhooks

| Tool | Description | |------|-------------| | get_webhook | Get your current webhook configuration | | set_webhook | Set a webhook URL for inbound events | | delete_webhook | Remove the webhook |

Environment Variables

| Variable | Required | Description | |----------|----------|-------------| | AGENTPHONE_API_KEY | Yes | Your AgentPhone API key | | AGENTPHONE_BASE_URL | No | Override the API base URL (defaults to production) |

Development

git clone https://github.com/AgentPhone-AI/agentphone-mcp.git
cd agentphone-mcp
npm install
npm run dev     # Run with tsx (hot reload)
npm run build   # Compile TypeScript
npm start       # Run compiled JS

How It Works

This is an MCP (Model Context Protocol) server that connects your AI assistant to the AgentPhone API. It runs as a local process that your AI client (Cursor, Claude Desktop, etc.) communicates with over stdio.

Your AI Assistant  <-->  agentphone-mcp (local)  <-->  AgentPhone API  <-->  Phone Network

The MCP server itself is stateless — it's a thin typed client that translates MCP tool calls into AgentPhone API requests. All state (numbers, calls, messages) lives on the AgentPhone platform.

License

MIT

MCP Servers