AgentPhonepricing, plans and limits

AgentPhone gives any AI agent its own US or Canadian phone number via 1 API call, with unified voice, SMS, real-time transcription, and built-in MCP support.

  • ai sdk
  • Web
checked

Last updated: 2026-07-01

Launched in 2026 by Y Combinator's Spring batch, AgentPhone is a telephony API that provisions a US or Canadian phone number for an AI agent through a single call. Voice and SMS both land on one webhook, with real-time call transcription and a native MCP server for Claude Code, Cursor, and Windsurf.

About AgentPhone

AgentPhone is a developer API launched in Spring 2026 by Meet Modi and Manav Modi as part of Y Combinator's Spring 2026 cohort. The platform solves a specific gap in the AI agent stack: AI agents that can handle web tasks lack the ability to make and receive phone calls or text messages, which are how most businesses still communicate. The core architecture handles both voice and SMS with an identical event shape, so developers write one webhook handler for all communication types. Incoming calls are transcribed in real time and delivered as text to the webhook; the agent replies with text, and AgentPhone converts the response to speech. Agent code never needs to handle audio processing, PSTN connectivity, or carrier compliance directly. The platform ships with a native MCP server, letting developers integrate directly from Claude Code, Cursor, Windsurf, Zed, and any generic MCP client without additional integration work. AgentPhone targets backend engineers and AI agent developers building applications where agents need to contact people by phone. Specific use cases include AI receptionists that handle inbound business calls, sales agents that follow up with leads by text, on-call engineering workflows where agents page engineers via SMS, coding agents that receive verification codes, and 2FA distribution systems. Early adopters include teams at Google's Agent Development Kit, Replit, LangChain, and Alchemy. Pricing is usage-based and not publicly listed; the team handles rates on request. Phone numbers are currently limited to US and Canadian numbers, with international coverage in progress. The platform supports Python and Node.js SDKs and a REST API. Area code selection is available during provisioning.

Pricing

Usage-based pricing, not publicly listed. The team quotes custom rates directly, covering number provisioning, voice minutes, and SMS usage as separate billed line items. No confirmed free tier.

Key Features

  • Unified Webhook for Voice and SMS: Voice calls and SMS messages arrive at one webhook endpoint with an identical event format, so a single handler covers all phone communication types.
  • Real-Time Call Transcription: Incoming voice calls are transcribed to text in real time before reaching the webhook, so agent code only ever processes plain text and never raw audio.
  • One-Call Number Provisioning: US and Canadian phone numbers are provisioned instantly via a single API call, with optional area code selection and no manual carrier setup required.
  • Native MCP Server: A built-in MCP server lets developers provision numbers and send messages directly from Claude Code, Cursor, Windsurf, Zed, or any compatible MCP client without writing API wrappers.
  • Python and Node.js SDKs: Official SDKs for Python and Node.js plus a REST API cover the most common agent runtime environments, with support for Google ADK, LangChain, and Replit confirmed.

Pros

  • MCP server ships out of the box, covering 6+ major AI coding environments including Claude Code and Cursor, so developers avoid writing any custom integration code.
  • Founders bring deep telephony context: Meet Modi built WhatsApp agent infrastructure for 280 million businesses at Meta, giving the platform credibility at scale.
  • One webhook handles both voice and SMS with the same event shape, cutting agent code complexity compared to managing separate voice and messaging APIs.

Cons

  • Coverage limited to US and Canadian numbers as of May 2026; any international agent workflow requires a separate telephony provider like Twilio or Bandwidth.
  • Pricing is not published, meaning developers cannot evaluate cost without a sales call, which slows down the typical developer-led evaluation cycle.
  • 2-person team (YC Spring 2026) means limited documentation, community support, and uptime SLA guarantees compared to established infrastructure providers like Twilio.

Frequently Asked Questions

What does AgentPhone actually cost?

Rates are quoted on request rather than listed, and billing is usage-based across number provisioning, voice minutes, and SMS. Teams sending business SMS in the US should also budget for A2P 10DLC registration, which runs $4-25 per brand, $10 per campaign, and roughly $0.003 per message in carrier surcharges. A quote comes from the team directly at agentphone.ai.

Can you use AgentPhone without paying?

There is no free tier and no trial, and usage is billed from the first call. The 2-person YC-backed team handles pricing through direct conversation instead of a self-serve plan page. Developers who want to prototype at no cost can start on Twilio's trial credits or Vapi's documented free developer tier.

What are AgentPhone's closest competitors?

The closest alternatives to AgentPhone are Retell AI, Vapi, and Twilio. Retell AI targets non-technical teams that want a no-code flow builder with call analytics and CRM hooks rather than raw webhooks. Vapi is API-first with public per-minute pricing and a free developer tier, which is easier to evaluate before committing than AgentPhone's sales-quote model. Twilio remains the broadest option, covering voice, SMS, WhatsApp, and email with global number coverage, at the cost of a steeper setup than AgentPhone's single endpoint.

What separates AgentPhone from Retell AI?

AgentPhone and Retell AI sit at different layers of the telephony-for-AI stack: AgentPhone is a bare-metal API that hands a phone number and webhook to your own agent code, while Retell AI is a managed platform with built-in conversation logic, latency tuning, and CRM integrations for Salesforce, HubSpot, and Zapier. Choose AgentPhone if you are writing custom agent logic and want minimal abstraction between your code and the carrier; choose Retell AI if you need a production-ready voice agent without building conversation handling yourself. Retell AI also publishes per-minute pricing ($0.07-0.11), which is easier to budget against than AgentPhone's sales-quote model.

How do you set up AgentPhone?

Four steps put an agent on the phone. Create an account at agentphone.ai and copy a key, then install the SDK (pip install agentphone or npm install agentphone) or point Claude Code or Cursor at the MCP server. Call the /numbers endpoint with a preferred area code to provision a US or Canadian number, then route that number to your agent's webhook, where calls arrive as live transcripts and SMS lands in the same event shape.

Top Alternatives

  • Retell AI: Retell AI ships a no-code voice agent builder with call analytics and CRM integrations; AgentPhone hands your own code a number and a webhook and stays out of the way.
  • ElevenLabs: ElevenLabs is the stronger voice synthesis engine once telephony is already solved, while AgentPhone bundles number provisioning and TTS behind one API.
  • Deepgram: Deepgram is standalone speech-to-text; AgentPhone covers the call itself, the number, and the transcription together.

More AI Tools on HokAI

Visit AgentPhone Official Website