# AgentCloud — Give your agent an iPhone

> Create, operate, and inspect a hosted iOS Simulator through one OAuth-enabled MCP URL.

AgentCloud connects every agent action to the evidence available at that moment: the screen, accessibility tree, application and Simulator logs, artifacts, and full-run video.

- Canonical page: [https://agentcloud.so/](https://agentcloud.so/)
- Create an account: [https://app.agentcloud.so/signup](https://app.agentcloud.so/signup)
- Log in: [https://app.agentcloud.so/login](https://app.agentcloud.so/login)

## Pricing

Monthly minimums become Simulator usage credit. Additional usage continues at the same plan rate.

| Plan | Monthly minimum | Active-second rate | Included Simulator time | Concurrent sessions | Evidence retention |
| --- | ---: | ---: | ---: | ---: | ---: |
| Developer | $49 | $0.00050 | About 27 hours | 1 | 7 days |
| Team | $199 | $0.00040 | About 138 hours | 4 | 30 days |
| Scale | $999 | $0.00033 | About 841 hours | 10 | 90 days |

Active time begins when the Simulator is ready and stops when the run is destroyed or reaches its inactivity timeout.

## How it works

1. **Authorize once in the browser.** OAuth 2.1 and PKCE connect an MCP client to the correct AgentCloud team without a manually managed agent secret.
2. **Create a disposable Simulator.** Install a Simulator-compatible `.app`, launch it, open deep links, observe the screen, and act through semantic or coordinate input.
3. **Review durable evidence.** The Simulator disappears at the end of the run, while actions, screenshots, accessibility state, logs, artifacts, and the finalized recording remain inspectable.

## Evidence retained with a run

- **Screen:** Visual state before and after actions.
- **UI tree:** Accessibility elements and semantic targets.
- **Logs:** Recent Simulator and application output.
- **Timeline:** Actions aligned to screenshots and video.
- **Artifacts:** Downloadable evidence retained by the team.

## Security and boundaries

- OAuth 2.1 with PKCE.
- No raw worker shell access.
- Disposable Simulator sessions.
- Browser sessions, OAuth grants, runs, and artifacts are scoped to the creating team.
- Evidence remains until its retention period ends or an authorized team member deletes it.

## Frequently asked questions

### What is an iOS Simulator MCP server?

It exposes a focused set of iOS Simulator operations as Model Context Protocol tools. A compatible coding agent can create a run, install and launch an app, observe the screen and accessibility tree, act on the interface, read logs, and destroy the session.

### Does the agent need to run on a Mac?

No. The iOS Simulator runs on an AgentCloud Mac worker while the MCP client connects remotely. The app must still be built for the iOS Simulator runtime.

### Can AgentCloud install an App Store IPA?

No. AgentCloud accepts Simulator-compatible `.app` bundles. A device-signed `.ipa` targets physical devices and is not interchangeable with a Simulator build.

### What does AgentCloud retain?

Runs can retain actions, screenshots, accessibility snapshots, logs, typed values, artifacts, and full-run video. This evidence may contain sensitive data, so teams should manage access and deletion deliberately.

### How does per-second billing work?

The monthly minimum is usage credit. Active Simulator time is totaled across sessions and billed at the plan's per-second rate. Usage beyond the included credit continues at the same rate.