A remote iOS control loop
AgentCloud is built for engineers whose coding agent runs on Linux, in CI, over SSH, or inside another cloud environment that cannot open a local Mac Simulator. The agent connects to one OAuth-enabled Model Context Protocol endpoint, receives a focused set of iOS tools, and works against a disposable hosted Simulator.
An authorized agent can upload build-ready iOS source or a Simulator-compatible .app, create a Simulator, install and launch the app, observe its screen and accessibility tree, tap, swipe, type, open deep links, and inspect application output. The control loop is designed for software verification, not general remote-computer access.
Evidence stays connected to the action
Every retained run ties actions to the state around them: screenshots, accessibility snapshots, application and Simulator logs, artifacts, and recording segments. A reviewer can follow what the agent saw, what it attempted, and what the app returned without reconstructing the session from unrelated screenshots and terminal fragments.
Simulator sessions are disposable, while evidence follows the retention policy for the workspace plan. Team boundaries apply across browser sessions, OAuth grants, runs, builds, and artifacts. Authorized workspace members can inspect shared evidence, and owners or admins can delete it where the control panel exposes that action.
A deliberate product boundary
AgentCloud does not provide a macOS desktop, an Xcode user interface, or a raw worker shell. It accepts Simulator-compatible application bundles, not device-signed .ipa files, and it does not replace physical-device testing. Network helpers are constrained to run-scoped tickets and configured port ranges rather than exposing the worker control plane.
Start with the developer resources, compare the available plans, or create a free workspace without a payment card. Product, security, privacy, and billing contact paths are listed on the contact page.