AgentPhone

Developers building AI agents that need to call or text people have to stitch together telecom primitives like Twilio with separate handling for voice and messaging, plus their own transcription and text-to-speech layers. This means building distinct pipelines for calls and messages, managing conversation threading manually, and integrating low-level telecom building blocks that were not designed with autonomous agents in mind. Use cases like sales follow-up calls, 2FA/verification handling, receptionist bots, on-call engineering flows, and coding agents all need an agent to reliably place calls and send texts without a human in the loop.

AgentPhone provisions US and Canadian phone numbers instantly via a single API call and routes both voice calls and messages through one unified webhook format. Incoming voice calls are transcribed in real time and delivered to the webhook as text; the agent replies with text and AgentPhone handles the text-to-speech conversion, so calls and messages share the same event shape. It offers a Python SDK, Node.js SDK, REST API, and a native MCP server, letting MCP-compatible agents like Claude Code, Cursor, OpenClaw, and Windsurf provision numbers and handle calls/messages through native tool use. The site shows example code for creating an agent, buying a number, and placing a call in a few lines.