Is ClawCall a direct alternative to AgentPhone?

They overlap but are different shapes. ClawCall is a finished outbound AI calling product that ships with a skill for Claude Code, Cursor, ClawHub, and OpenClaw — you call the hosted REST API at api.clawcall.dev and can retrieve captured transcript evidence when present and temporary recording access when available. AgentPhone is YC-backed infrastructure that gives every AI agent its own US or Canadian phone number and routes inbound voice and SMS to one webhook. If you want an agent that places calls today, ClawCall is closer. If you want each of your agents to own a phone number for inbound traffic, AgentPhone is closer. Switching between them is mostly rewriting your handler — there is no shared portable agent format.

Which one supports inbound calls?

Both do. AgentPhone sends inbound voice and SMS to your agent's number to your webhook as text and converts your text replies back to speech, and it can also place outbound calls. ClawCall makes outbound calls on every paid plan and offers focused inbound reception on the Unlimited Reserve Plus tier at $14.99/mo, which adds an AI inbound assistant on one active reserved number. For multi-tenant inbound at scale, AgentPhone's per-agent-number model is the more natural fit; for one consumer or developer who wants a single inbound number alongside outbound, Reserve Plus is enough.

Can I use either inside an AI coding agent like Claude Code or Cursor?

ClawCall's preferred agent connection is hosted OAuth MCP at https://api.clawcall.dev/mcp; compatible hosts sign in and discover the calling tools without receiving an API key. The packaged skill and REST API are fallbacks for hosts without remote MCP and OAuth. On the REST fallback, the first anonymous POST /call returns an API key you can keep using, and it carries over when you sign up. AgentPhone offers SDKs, a REST API, an MCP server and an MIT-licensed Claude Code skill, so a coding agent can provision numbers and place calls, but your agent still supplies the conversation logic.

Does either guarantee the AI will disclose itself or avoid voicemail?

ClawCall enforces it as a product rule: the agent always discloses it is an AI when asked, can leave voicemail when instructed, and does not send bulk or robocall-style promotions, while individual outreach calls are allowed. That is a non-negotiable platform-level behavior, not a prompt suggestion. AgentPhone is infrastructure — your agent's prompt controls its behavior, so disclosure and voicemail policy are your responsibility. That is appropriate for a number-provisioning layer, but it means consumer-facing builders on AgentPhone need to encode those behaviors themselves and own the regulatory and brand risk that comes with them.

Which one should I pick if I just want my AI to call the dentist?

ClawCall. The use case it is explicitly built for is consumer chores in the US: doctor and dentist appointments, restaurant reservations, disputing medical and utility bills, canceling subscriptions, airline rebookings, DMV, insurance, and eliminating hold time. The no-card trial of 30 calls and 30 minutes, whichever lasts later lets you confirm it works before committing to a plan from $4.99 a month. AgentPhone is built for teams shipping agents that need their own phone identity for inbound calls and SMS, not for the one-off consumer outbound task.

Comparison

ClawCall vs AgentPhone

AgentPhone is YC-backed infrastructure that gives each AI agent its own number for calls and texts; ClawCall is a finished agent that places US calls for you.

Try a free call Visit AgentPhone

The short answer

AgentPhone is a Y Combinator-backed API that gives each AI agent its own US or Canadian phone number, sends inbound calls and texts to one webhook as text, speaks your agent's text replies, and can place outbound calls from a hosted voice agent through SDKs, a REST API or an MCP server; its site does not publish prices. ClawCall is a finished AI phone agent that calls US numbers in English for $4.99-$14.99 a month flat: you give it a number and a task, and it works through the phone menu, waits on hold, talks to whoever answers and can loop you in, after a free trial of 30 calls and 30 minutes, whichever lasts later.

Pick AgentPhone if you are building a product where each agent needs its own number for inbound calls, texts and verification codes, and you want to write the agent's logic; pick ClawCall if you want an agent that already knows how to call businesses, work through menus and holds, and loop you in, behind one sign-in.

Connect ClawCall to ChatGPT, Claude, Grok Bot or another AI agent, or call it from your own code. See how AI agents connect.

Feature comparison

Check AgentPhone details on the vendor’s own site before you buy.

Feature	ClawCall	AgentPhone
Primary shape	Finished AI calling product + drop-in agent skill	Developer API for phone-number infrastructure
Core promise	Your agent makes a real outbound call and returns captured transcript evidence when present + temporary recording access when available	Your agent gets a phone number; inbound voice and SMS arrive at one webhook
Pricing model	Flat monthly: $4.99 / $8.99 / $14.99; 30 calls and 30 minutes, whichever lasts later, no card	Not published
Number geography	US only (+1 NANP)	US and Canada; more countries planned
Inbound vs outbound	Outbound by default; inbound on Unlimited Reserve Plus tier ($14.99/mo)	Both: inbound voice and SMS to one webhook; outbound calls through calls.create with a hosted voice agent
Integration surface	Hosted OAuth MCP at api.clawcall.dev/mcp (preferred); REST API + packaged skill fallbacks	Python and Node SDKs, REST API, MCP server and a Claude Code skill
SMS	SMS/iMessage user-facing interface; no outbound SMS via public API	Inbound SMS handled in the same webhook as voice
Voice stack	Telnyx + LiveKit Agents: Meta Muse STT, Claude, ElevenLabs v4 Turbo (managed)	AgentPhone handles TTS; inbound voice sent to your webhook as text
AI-honesty rule	Always discloses it is an AI when asked; can leave voicemail when instructed	Unspecified (your agent's prompt controls behavior)
Auth model	OAuth for hosted MCP; the REST fallback returns an API key on the first anonymous call	Unspecified
Open source / community	Docs under CC BY 4.0; agent skill open	Claude Code skill is MIT-licensed
Backing	Independent	Y Combinator-backed

Which one fits

Choose ClawCall if…

You want an AI coding agent (Claude Code, Cursor, ClawHub, OpenClaw) to place real outbound phone calls today via a drop-in skill — not wire up a webhook.
You need an end-to-end outbound calling product (dial, IVR navigation, hold, captured transcript evidence when present, and temporary recording access when available) rather than phone-number infrastructure you compose yourself.
You prefer flat monthly pricing ($4.99, $8.99, or $14.99 a month) with a no-card trial of 30 calls and 30 minutes, whichever lasts later over per-minute telecom-style billing.
You care about an enforced AI-honesty rule and instruction-controlled voicemail / no-robocall posture for consumer-facing calls.
You want SMS and iMessage as a user-facing channel into the same agent, not just as a developer webhook event.
Your use cases are consumer chores in the US: appointments, reservations, bill disputes, cancellations, airline rebooking, DMV, hold-time elimination.

Choose AgentPhone if…

You are building a multi-tenant agent platform where each agent needs a persistent, dedicated phone number for inbound calls and SMS — AgentPhone is purpose-built for that identity layer.
You need Canadian numbers in addition to US numbers at launch; ClawCall is US-only today.
You want a single unified webhook event shape for both inbound voice and SMS so your agent code stays uniform across channels.
You value Y Combinator backing, a public GitHub org, and a Discord community for an early-stage infrastructure vendor.

What do you get on day one with each?

The clearest way to separate these two is to ask what arrives in your inbox the day after you sign up. With ClawCall, what arrives is a working agent: you POST /call with a US number and an instruction, get a call_id back immediately, poll GET /call/:id until the call is finalized, and read captured transcript evidence when present and temporary recording access when available.

The voice stack is already glued together and managed: Telnyx for telephony, LiveKit Agents for realtime turn-taking, Meta Muse for speech recognition, a Claude model for the conversation, and ElevenLabs v4 Turbo for voices. The agent skill installs into Claude Code, Cursor, ClawHub, or OpenClaw, so a coding agent can place a call in the same session it writes code.

With AgentPhone, what arrives is a phone number and a webhook contract: inbound voice is transcribed in real time and pushed to your webhook as text, your agent replies with text, and AgentPhone runs the text-to-speech layer using the same event shape for both voice and messaging. You still own the agent's brain, its prompt, its tools, and its memory. That is the right shape if you are shipping a platform where every tenant agent needs its own persistent number; it is more work than you want if your goal is simply to have an AI call the dentist.

What does each one cost?

ClawCall publishes three flat monthly plans — Unlimited at $4.99/mo (unlimited calls from a shared outbound number pool), Unlimited Reserve at $8.99/mo (adds one private reserved caller-ID number), and Unlimited Reserve Plus at $14.99/mo (adds an AI inbound assistant on that reserved number) — plus a free trial of 30 calls and 30 minutes, whichever lasts later that does not ask for a credit card.

There is no per-minute meter, no annual contract, and legacy minute-pack purchases were retired. The implicit bet is that a consumer or solo developer wants to know the bill in advance and not babysit a usage dial. AgentPhone's public site emphasizes that it is built for teams shipping agents and provisions numbers instantly through the API, but pricing is not published on its site, so the table lists it as not published.

Confirm rates with AgentPhone before you plan around them. If predictable flat pricing matters more than elasticity, ClawCall's posture is friendlier.

Integration: hosted OAuth MCP vs unified webhook

Both products want to live inside an AI agent's tool belt, but they hand you different primitives. ClawCall's preferred primitive is its OAuth-protected hosted MCP endpoint at https://api.clawcall.dev/mcp. Compatible hosts sign in, approve calls:read and calls:write, and discover twelve bounded tools without receiving a ClawCall API key. The packaged skill and REST API remain fallbacks for hosts without remote MCP and OAuth; the REST fallback returns an API key on the first anonymous POST /call, and that key carries over when you sign up.

AgentPhone exposes a REST API, an MCP server, and an NPM package, and routes both inbound voice and inbound SMS into one webhook event shape — so an agent's handler code looks the same whether the user texted or called. That uniformity is genuinely useful when you are building a product where every agent is a long-lived entity with its own number. Pick by whether you need ClawCall's finished outbound workflow or AgentPhone's per-agent number and unified inbound webhook.

Geography, channels, and the AI-honesty rule

AgentPhone provisions US and Canadian phone numbers instantly through the API and tells customers more countries are coming soon. ClawCall is US-only today (+1 NANP) and English only. Every plan allows one outbound call at a time per account. On channels, AgentPhone treats inbound voice and inbound SMS as siblings in one webhook; ClawCall offers SMS and iMessage as a user-facing interface into the agent but does not expose outbound SMS through its public API.

The hard ClawCall differentiator that does not appear in AgentPhone's marketing is behavioral: ClawCall always discloses it is an AI when asked, can leave voicemail when instructed, and does not send bulk or robocall-style promotions, while individual outreach calls are allowed. For a consumer-facing calling agent, that posture is a brand and compliance asset; for an infrastructure vendor it is correctly out of scope, because the agent's prompt and behavior are the customer's responsibility, not the platform's.