AgentPhone is a communication infrastructure provider for the AI agent stack. It is primarily relevant to developers building autonomous agents that need to cross the barrier between digital environments and the legacy telephone network. By providing a phone identity to an agent, AgentPhone enables workflows such as autonomous 2FA verification, outbound sales calls, and real-time support reception.
They are active participants in the Model Context Protocol (MCP) ecosystem. By offering an MCP server, they allow agents within developer environments to use telephony as a standard tool. This aligns them with the shift toward "tool-use" as the primary way agents interact with external services. For people building agents that need to be reachable or perform tasks over the phone, AgentPhone simplifies the transcription and connection logic into a single developer interface.
The most capable AI agents are currently limited by their inability to interact with the world outside of a browser or an API. While Large Language Models (LLMs) can write code or draft emails, they are often cut off from the phone network—a critical piece of infrastructure for everything from customer follow-up to two-factor authentication. AgentPhone is an infrastructure provider based in San Francisco that bridges this gap by giving AI agents their own dedicated phone numbers for voice and messaging.
Founded by Manav Modi and Meet Modi, and backed by Y Combinator, the company is part of a growing wave of "agent infrastructure" startups. These firms build the plumbing required for AI to act as an autonomous worker rather than just a chatbot. AgentPhone specifically targets developers shipping agents that need to handle sales leads, operate as virtual receptionists, or manage on-call engineering rotations where a phone call is the standard escalation path.
The core of the service is a unified API that simplifies how agents receive data. In a typical telephony setup, a developer would have to manage separate pipelines for SMS and voice, often using different libraries to handle audio streaming and speech-to-text. AgentPhone abstracts this by using a single webhook format. When an agent receives a voice call, the system transcribes the audio in real time and sends the text to the developer's webhook. The agent can then respond with text, which the platform converts back into speech for the caller. This approach allows developers to treat a phone call essentially as a long-form text message, significantly lowering the barrier to building voice-capable agents.
One of the most distinct technical choices AgentPhone has made is the adoption of the Model Context Protocol (MCP). This allows the service to run as an MCP server, which means agents running in environments like Claude Code, Cursor, or Windsurf can use a phone number as a native tool. Within these clients, an agent can programmatically provision a number, send a text, or check for an incoming 2FA code without the developer writing custom integration logic for each project.
This level of tool-readiness separates them from incumbents like Twilio. While Twilio is a massive utility for telecommunications, its complexity is a friction point for AI developers who want to give an agent a "cell phone" identity in a few lines of code. AgentPhone includes features like iMessage-enabled numbers and RCS fallback, targeting the specific communication standards that human users expect when interacting with an entity that has a phone number.
AgentPhone operates on a usage-based pricing model, starting at $3 per number per month for basic usage. For more advanced needs, including iMessage support, they offer a $150 per month Pro tier. This pricing structure reflects their focus on the developer and startup market, providing a self-serve path for buying up to 10 numbers instantly. They also offer free number porting from other providers, clearly aimed at capturing users frustrated by the overhead of legacy providers. As agents become more prevalent in corporate workflows, the ability for an AI to have a persistent, recognizable phone identity becomes less of a novelty and more of a requirement for reliable automation.
Programmable phone numbers for AI agents with voice transcription and messaging.
AgentPhone is hiring.