PhoneUse: Cloud Phones and Real iPhones for AI Agents
PhoneUse gives AI agents a phone. It rents cloud Android phones that boot in about four seconds, plus real retail iPhones in a San Francisco rack that a robot stylus taps while an overhead camera reads the screen. An agent drives either kind through the same tools: an MCP server you add to Claude, Cursor, Codex or Windsurf, an HTTP API, or a command-line tool. It’s built for the jobs that only happen in phone apps, like posting to Instagram, ordering from delivery apps with no website, receiving a real text message, or running app tests across many devices. PhoneUse is run by Masta, Inc. and is for business use only.
The quick answer: PhoneUse is part of a new wave of companies selling phones as infrastructure for AI agents, and its documentation is unusually candid about what’s live and what isn’t. Today, the phones you can drive through its API are cloud Android devices. The robot-tapped iPhone is sold separately in its console and isn’t part of the API yet. A disclosure: we make Carly, an AI assistant that makes and answers phone calls and texts. That’s a different job from what PhoneUse does, which is let an agent operate a phone’s apps. Everything below comes from PhoneUse’s site, docs and terms.
PhoneUse at a glance
Checked against PhoneUse’s site, documentation, terms and acceptable use policy on November 12, 2026.
| PhoneUse | |
|---|---|
| What it is | Phones for AI agents: cloud Android devices and robot-operated iPhones |
| How agents connect | A hosted MCP server, an HTTP API and a CLI on npm |
| Works with | Any MCP client, including Claude, Cursor, Codex and Windsurf |
| Phones on the API today | Cloud Android phones, billed monthly or by the minute |
| Real iPhones | Bought in the console, prepared within 7 days, not yet in the API |
| Who can use it | Businesses and professionals, 18 and over |
| Company | Masta, Inc., San Francisco |
What PhoneUse is for
PhoneUse’s homepage starts from a simple observation: most of life runs on phone apps, and agents stall “the moment a task needs a native app, a real text message or a phone the app trusts.” Browser agents can’t leave the web, and emulators on a laptop don’t scale. PhoneUse’s answer is a fleet of phones an agent can rent.
The use cases it lists:
- Social and marketing. Posting, replying and growing accounts on TikTok, Instagram and Reddit. Its demo shows an agent replying to every comment on an Instagram Reel.
- Mobile-only apps. Ordering on apps such as Swiggy, Zomato or Uber that have no web version an agent could reach.
- Real phone numbers. A phone with a SIM receives verification codes, “so flows never stall at login.”
- Proactive texting agents. Agents that text first, “with the full context of a real phone behind them,” like messaging a customer when an order ships.
- Testing and QA. Running test suites across hundreds of Android and dozens of iOS devices in parallel.
- Agent sandboxes. Clean phones where agents can learn and be evaluated before production.
Cloud phones and real iPhones
PhoneUse sells two very different kinds of device.
Cloud phones are virtual Android devices that boot in about four seconds and scale to hundreds. You pick a country when you buy one, because a cloud phone comes up on that country’s mobile network, which decides which app stores and services treat it as local. A “Real phone identity” version costs more.
Real iPhones are retail iPhones and Androids racked in San Francisco. The homepage describes a retail iPhone 15 Pro with a real carrier SIM: “No jailbreak, no root and no helper app.” A stylus touches the glass, an overhead camera reads the screen, and the phone is wired and always charged. The point is that apps can tell when they run on a virtual device, and a real phone touched by a physical stylus “looks exactly like a person.”
The homepage presents switching between the two as changing one parameter in your code. The documentation is more precise about where things stand: every device you can reach through the API today is a cloud phone, and the robot hand with an iPhone “is not part of the API, MCP or CLI.” You buy it in the console, and the team prepares it within seven days of payment.
How an agent drives a phone
PhoneUse calls itself “the mobile execution layer for AI agents.” An agent works through a small set of tools on its hosted MCP server: list and acquire a device, take a screenshot or read the screen as a tree of named elements, then tap, swipe, type, press keys, open apps and release the device. Every device uses the same coordinate grid, from 0 to 1,000 on both axes, so the same tap works on any screen size.
A few design choices stand out in the docs:
- Look before you touch. An action is refused until the agent has fetched at least one frame of the screen, “because coordinates with no frame behind them have no meaning.”
- Evidence for every run. Actions report when the device actually committed them, and a retry that reuses its key replays the recorded result instead of acting twice.
- Refusals, not silent failures. “An action a device cannot perform is refused. It is never accepted and quietly dropped.”
- Leases. Acquiring a device gives you a lease that expires after 10 minutes unless renewed, and no lease lasts longer than 60 minutes, so a crashed agent frees the phone on its own.
There are two ways to work. You can drive it yourself, with your agent deciding every gesture, or hand over a goal in a sentence and let an executor run until it succeeds or gives up. PhoneUse can run that executor itself, or you can bring your own worker. Runs carry hard limits: a default of 50 steps, which you can only lower, and a deadline.
PhoneUse also has a copilot in its console that lets you chat with a phone and watch it work. Its acceptable use policy notes that the copilot currently sends instructions to an OpenAI model through OpenRouter.
When a person has to step in
Some screens are meant to stop automation, like a sign-in, a one-time code or a challenge, and PhoneUse’s docs say plainly that “an agent should not be trying to defeat them.” Instead, PhoneUse asks the person running the agent to go and handle the screen on the device, then confirm when they’re done.
The docs are firm about passwords: “PhoneUse asks a human to go and type a password into the phone in front of them; it never asks anyone to send it one.” If the person declines, cancels or times out, the request fails closed.
Pricing
PhoneUse’s billing docs, with rates checked October 18, 2026, list:
| Phone | Price |
|---|---|
| Cloud phone, monthly | $20, $24, $28 or $32 a month, by model |
| Cloud phone with real phone identity | $39 a month |
| Cloud phone, per minute | $0.024 a minute from a prepaid balance |
| Robot hand + iPhone | $149 a month per robot, bought in the console |
| Enterprise | Priced on a call, with dedicated racks, SSO, audit logs and an SLA |
Monthly phones have no meter or overage. Per-minute phones charge from a prepaid balance, so you can’t spend more than you’ve added, and PhoneUse emails you about 10 minutes before the balance runs out. Its docs also state, unusually directly, that “there is no dollar limit you can set on this surface today,” and that a per-minute phone keeps charging while it’s rented to you whether or not anything is using it.
Rules on what you can do with it
PhoneUse’s marketing and its rules pull in slightly different directions, and anyone evaluating it should read both.
The homepage sells “super undetectability” and shows its real iPhones passing app-side checks for emulators, automation apps and device integrity. Founder Abdulla Ababakre told RuntimeWire in October 2026 that “the big platforms never detect it.”
PhoneUse’s acceptable use policy then sets three rules on top of “nothing illegal”: no fraud, including fake engagement and fabricated identities; no taking accounts that aren’t yours; and no getting around a platform’s own protections, naming rate limits, device attestation, bot detection, region locks and account-linking protections. It explains the tension itself: a real handset with a real number “is exactly the shape of thing those protections exist to see. Having a real device does not make you authorised to use it against somebody else’s account.”
In practice, PhoneUse fits best where you’re operating your own accounts and apps, testing software, or doing tasks the app’s own terms allow. It reserves the right to throttle, suspend or terminate access when the rules are broken, and suggests emailing the team before building anything you’re unsure about.
Data and screenshots
Screenshots of a phone in use can contain anything on the screen, and PhoneUse’s frame retention page treats them as “the most sensitive thing the control plane touches.” Worth knowing before you start: when frame storage is configured, the default policy keeps archived frames with no expiry, and the service doesn’t automatically delete them. “Do not assume that returning a device, deleting a chat thread or closing an account erases a separately archived frame.” Organizations can set their own retention policy, and credentials are scoped so a key can only see frames from devices it’s allowed to touch.
The company
PhoneUse is run by Masta, Inc., a San Francisco startup founded by Abdulla Ababakre, who is also the contact for demos and acceptable use questions. Its CLI is published on npm as @phoneuse/cli, and its public changelog shows frequent releases through October 2026. PhoneUse is self-serve: you can sign up, buy a cloud phone, drive it from a browser and create an API key without talking to anyone.
How PhoneUse compares
PhoneUse sits in a new category of phone-use infrastructure for agents, alongside Glasshand, which rents dedicated real iPhones and Macs by the month, and the open-source Droidrun framework, now called Mobilerun. It’s closest in spirit to browser infrastructure such as Browserbase and Browser Use, applied to phones, which we cover in our Browser Use alternatives guide. Its distinguishing features are the physical iPhone rack and how carefully its docs describe what each part does.
If what you actually need is phone calls and texts rather than an agent tapping through apps, Carly does phone and SMS out of the box: it makes and answers calls and texts from a real business number, and handles email and calendars too, from inside ChatGPT or Claude or by text.
Who PhoneUse suits
PhoneUse is a good fit if:
- You’re building an agent that has to use mobile-only apps or receive real text messages.
- You run mobile app tests and want many Android devices on demand.
- You want MCP access to a phone from Claude, Cursor or Codex without setting up emulators.
- You’re operating your own accounts and want a device that apps treat like a phone.
Look elsewhere if:
- You need real iPhones driven through an API today; that part isn’t live yet.
- Your task is mostly on the web, where a browser agent is simpler.
- You’re a consumer. PhoneUse is for business and professional use.
- Your plan depends on getting around another service’s bot detection, which its rules forbid.
FAQ
What is PhoneUse?
PhoneUse is a service from Masta, Inc. that rents AI agents phones to operate: cloud Android devices and real iPhones tapped by a robot stylus. Agents drive them through an MCP server, an HTTP API or a CLI.
How much does PhoneUse cost?
Cloud phones cost $20 to $32 a month by model, $39 a month with a real phone identity, or $0.024 a minute from a prepaid balance. A robot hand with a real iPhone costs $149 a month. Enterprise is priced on a call.
Does PhoneUse work with Claude?
Yes. PhoneUse’s MCP server works with any MCP client, including Claude, Cursor, Codex and Windsurf, using an API key.
Can PhoneUse control a real iPhone?
PhoneUse sells real iPhones operated by a robot stylus with an overhead camera. Its docs say the robot hand isn’t part of the API, MCP or CLI yet, and is bought and prepared through the console.
Can I use PhoneUse to avoid bot detection?
No. PhoneUse’s acceptable use policy forbids using its phones to get around another platform’s protections, including bot detection and device attestation, as well as fraud and taking over accounts that aren’t yours.
Ready to automate your busywork?
Carly schedules, researches, and briefs you—so you can focus on what matters.
See what people say
"Before Carly, I relied on a Calendly link, but the whole process felt impersonal and not very professional. Carly changed that by handling all the back-and-forth, so I'm no longer stuck in endless email threads trying to line up schedules.
Now Carly reaches out to candidates, shares my real-time availability, lets them pick a slot, then sends a Zoom link and drops it straight into my calendar. She sends reminders to both of us before each call, which has significantly reduced no-shows and last-minute confusion.
On top of scheduling, Carly acts like a full executive assistant, sending me my schedule the night before so I can prepare for each call. It reminds me of the old x.ai assistant, but Carly is noticeably smarter, faster, and better suited to my healthcare recruitment business."
