---
title: "FAQ, MCP & limits"
section: "Agent Pool & Device Farm"
description: "Answers to common questions about SmartestQA test agents and Device Farm, how AI assistants use agents through MCP, and the module’s limits."
url: https://docs.smartestqa.ai/en/agent-pool-faq.html
audience: [customer, admin]
updated: 2026-11-08
keywords: "agent faq device farm faq share agent api key rotate ios desktop windows macos mac2 live control reservation app library limit 30 mcp smartest_context smartest_run_case runOn agent limits"
---
# Agent Pool & Device Farm: FAQ, MCP & limits

Answers to common questions about test agents and Device Farm, what AI assistants can do here, and the limits.

## In short

**On this page you can**

- Find quick answers about agents, devices, live sessions and the App Library.
- Filter the questions for customers or admins.
- See what an AI assistant connected through MCP knows about your agents.
- Look up every limit and time-out of the module in one table.

**How to use it**

1. Pick an audience with the filter buttons above the questions.
2. Open a question; follow its link for the step-by-step page.
3. Check the limits table before you plan devices or apps.

## Questions and answers

### Q: Do I need to install an agent?

Not for a web app reachable from the internet: tests run on the Smartest Server. You need one for apps inside your network, to watch the browser, and for mobile and desktop apps on your own devices. See [Do you need an agent?](https://docs.smartestqa.ai/en/quick-start-connect.html#agent)

_Audience: customer_

### Q: Can my team share one agent?

Yes. Create a shared agent with **Agent** ([Create a shared agent](https://docs.smartestqa.ai/en/agent-pool-agents.html#new-agent)); everyone in the company sees it under [**Virtual Machine List**](https://docs.smartestqa.ai/en/agent-pool-agents.html#agent-list). Localhost agents are personal: you see only your own.

_Audience: customer, admin_

### Q: How many tests can one agent run at the same time?

As many as its worker limit, shown as the second number of [**Workers (active/total)**](https://docs.smartestqa.ai/en/agent-pool-numbers.html#host-kpis). More tests wait in **Queued**. See [Agent performance cards](https://docs.smartestqa.ai/en/agent-pool-numbers.html#host-kpis).

_Audience: customer_

### Q: What happens if my laptop with the agent goes to sleep?

After 90 seconds without a heartbeat the agent turns Offline, and runs that target it fail with “The following agent(s) are currently offline”. Use a machine that stays on, or the Smartest Server for public web apps. See [When is an agent Online?](https://docs.smartestqa.ai/en/quick-start-connect.html#agent-online-rule)

_Audience: customer_

### Q: Which operating system does the agent need?

Windows, macOS or Linux for web and Android. Desktop apps need an agent on the same operating system as the app: Windows apps a Windows agent, macOS apps a Mac agent. iOS devices and simulators need a Mac. See [Download Client](https://docs.smartestqa.ai/en/agent-pool-agents.html#download-client) and [Mac for desktop tests](https://docs.smartestqa.ai/en/agent-pool-local-setup.html#macos-desktop).

_Audience: customer_

### Q: What network access does an agent need?

The agent connects out to SmartestQA and keeps that connection open; your app must be reachable from the agent’s machine. A proxy or firewall that blocks the connection keeps the agent Offline. Only results (logs, [screenshots](https://docs.smartestqa.ai/en/report-steps.html#screenshots), video) go back to SmartestQA; see [Where do my tests run?](https://docs.smartestqa.ai/en/quick-start-faq.html#faq-where-run)

_Audience: admin_

### Q: How do I replace an agent’s API key?

There is no “new key” button. For a shared agent, delete it and create a new one, then start the client with the new key and update [pipelines](https://docs.smartestqa.ai/en/schedule-create.html) that targeted the old agent. Localhost agents cannot be deleted.

_Audience: admin_

### Q: Who can create, rename and delete shared agents?

Any user of the company can create shared agents, rename them and delete them. Agree on names in your team and do not delete agents that [pipelines](https://docs.smartestqa.ai/en/schedule-create.html) use. Deleting disconnects a running client at once.

_Audience: admin_

### Q: Can we connect our own phones to the Device Farm?

Yes, with the Device Farm right. An admin creates a node and runs its install command on the computer the phones are attached to. Devices on your own nodes are reserved for your company. See [Add a node](https://docs.smartestqa.ai/en/agent-pool-devices.html#create-node). Without the right, connect phones to your own agent instead ([Local device setup](https://docs.smartestqa.ai/en/agent-pool-local-setup.html)).

_Audience: admin, customer_

### Q: Can I keep a farm device for myself?

While you hold it with [**Take device**](https://docs.smartestqa.ai/en/agent-pool-live-device.html#farm-device), nobody else and no run can use it. [Admins](https://docs.smartestqa.ai/en/company-settings-roles.html) can mark devices **Manual** so automatic runs never pick them. Booking devices ahead of time is not offered in the app.

_Audience: customer_

### Q: Can I control an iOS device live?

Not yet. You can watch the screen; taps and keys work on Android only. See [What you can do in a session](https://docs.smartestqa.ai/en/agent-pool-live-device.html#controls).

_Audience: customer_

### Q: Do I upload my app again for every new build?

Use [**Replace app**](https://docs.smartestqa.ai/en/agent-pool-app-library.html#replace) on the existing app. The app keeps its place in [pipelines](https://docs.smartestqa.ai/en/schedule-create.html) and the new build is installed before the next run. See [Upload a new version](https://docs.smartestqa.ai/en/agent-pool-app-library.html#replace).

_Audience: customer, admin_

### Q: How many apps can the App Library hold?

30 apps per company, each with one stored build.

_Audience: customer, admin_

### Q: What changes between the cloud service and an on-premise installation?

On-premise, users also see the Nodes tab (read only), live metrics and device logs, and may install App Library builds on devices; admins get a read-only [Licence](https://docs.smartestqa.ai/en/company-settings-billing.html#licence-status) tab, and concurrency, device-minute and reservation limits do not apply. In the cloud service, users see the devices made available to their company and can drive only devices assigned to it. Full table: [Who can do what](https://docs.smartestqa.ai/en/agent-pool.html#who-sees-what).

_Audience: admin_

### Q: Are farm runs limited?

In the cloud service your [licence](https://docs.smartestqa.ai/en/company-settings-billing.html#licence-status) can limit how many farm runs your company has at the same time, which platforms and device kinds (emulators only or real devices) you may use, and device minutes. Minutes are counted per started minute ([Run minutes](https://docs.smartestqa.ai/en/agent-pool-numbers.html#run-minutes)). For your values, ask info@smartestqa.ai. On-premise these capacity limits do not apply.

_Audience: admin_

## For AI assistants (MCP)

An assistant connected through the SmartestQA [MCP server](https://docs.smartestqa.ai/en/mcp.html) (set up in the [MCP tab](https://docs.smartestqa.ai/en/agent-pool-agents.html#mcp-tab), using the selected agent’s API key) knows about your agents but does not manage them:

- `smartest_context` reports the environment, including “agents *n*/*m* online”, and warns when no agent is online and no server runner exists.
- `smartest_run_case` runs a scenario with `runOn` = `server` or `agent`. It refuses `agent` when none of your agents is online (`AGENT_OFFLINE`), runs desktop scenarios only on an agent, and never switches where a run happens without asking you.
- There is no MCP tool for nodes, devices, live sessions or the App Library today.

Example prompts

```text
Is any of my agents online in project "Mobile Shop"? If not, tell me what to start.

Run the scenario "Login with valid user" on my agent, not on the server.

Why did the last run fail: was it the app, or the agent being offline?
```

## Limits

| Item | Limit |
| --- | --- |
| Agent Offline after | 90 s without heartbeat (checked every 30 s) |
| Node Offline after | 180 s without heartbeat, or 60 s with the connection closed (checked every 20 s) |
| Agent and node names | Unique within the company |
| Agent performance history | 24 h, about one point per minute |
| Apps in the App Library | 30 per company, one build each |
| App file types | .apk, .aab, .ipa, simulator .zip |
| Recent runs loaded | 200 (cloud users: 100) |
| Rows per page in Recent runs | 10 · 25 · 50 · 100 |
| One farm run | 120 min, then closed as Failed |
| App install on a device | 6 min, then “timed out” |
| Live session lease | 180 s, renewed while open |
| Live reconnect attempts | 3 (after 4 s, 10 s, 20 s) |
| Desktop tests | Agent on Windows (Windows driver) or macOS (Mac2 driver) |
| iOS devices | Agent or node on macOS; live view without control |
