CLOUSD
Open account
Cloud PhonesDevicesCloud Android phones for your agentsObservabilityWatch the agent live: screen, screenshots, UI treeRegionsDevices online in 45 countriesRent a PhoneCloud Android phones by the minuteFree Cloud PhoneYour first 30 minutes are free
For DevelopersVirtual AssistantSoonNo-code assistant that runs phone tasks for youAI Agents & MCPHosted MCP and SDK for Claude, Codex and CursorAndroid Emulator OnlineAndroid in your browser, nothing to installRemote Android PhoneA cloud phone you control from any PC
Open SourcePython SDKScript devices and agents in a few linesMCP ServerOne command to plug devices into any MCP clientAgent SkillTeach your coding agent to run phones
Start Free Trial
Built ForAI Agent BuildersA cloud phone for every agent or userAI EmployeesA work phone for every AI workerQA TeamsTest your app on the phones people actually own
IndustryCommercePrices, promos and checkout per marketGamingTest builds and evaluate game agentsIT & SecuritySecurity testing for Android appsComplianceWhat apps show users in each country
See All Use Cases
LearnBenchmarkHow agents score on Clousd phonesBlogGuides, deep dives and field notesCompareClousd next to the alternatives
BuildIntegrationsSoonMCP, Claude Code, Codex, Cursor, n8nGitHubSDK, MCP server and examples
HelpServer StatusLive checks with 90 days of historyContact SupportA person answers within a working day
Read the Docs

Cloud Android phones for your AI agents.

Android phones in the cloud for AI agents to train, test and work on. Your agent looks at the screen, acts on it and reports back.

Get an API key Read the docs
One steplook · act · check
# 1. look: fresh screenshot + every element with its bounds
GET  /v1/devices/c3f9a41d2/observe
     → { "seq": 41, "image": {…}, "activity": "…Cart",
         "ui": [ { "text": "Checkout", "b": [60,1560,1020,1680] } ] }

# 2. act, and wait until the screen settles
POST /v1/devices/c3f9a41d2/act
{ "op": "tap_text", "text": "Checkout", "seq": 41, "settle": true }

# 3. check from inside
adb shell dumpsys activity top   # did the order go through?

An interface an agent can drive.

One call to look: a fresh screenshot and every element on the screen. One call to act, and it waits until the screen settles. ADB when a check needs to go deeper.

MCP serverPython SDKGitHub
Agent APIREST · JSON
POST/v1/runsA task in plain words
GET/v1/devices/{name}/observeScreen and UI tree
POST/v1/devices/{name}/actAct, wait to settle
GET/v1/devices/{name}/screenshotA screenshot
ADBper-device relayCheck from inside
Stale actions are refused, never guessed409 stale

A whole goal in one call.

An agent does not have to tap through every screen. It gives the phone the goal in plain words: Clousd's own agent does the steps and comes back with the answer and a replay of every step. One call can give the same task to many phones.

One taskMCP · Python
# a smoke test on two phones
run task="Open com.example.app and tell me what the first screen shows"
    devices="Pixel QA, Galaxy QA"
→ completed · Pixel QA: "Welcome, Sign in, Create account" · Galaxy QA: …

# or in Python
c.run("Open the app and report the first screen", device=phone, wait=True)

What teams use it for.

Agent evaluation

Test an agent on everyday apps and popular phone models.

Training environments

Repeatable runs across phone models and Android versions.

Agents that need a phone

Products that book, order and message for their users.

Testing with agents

An agent walks through your app on cloud phones. App testing

Start in a minute.

  1. Create an accountEmail and password, or Google or GitHub.
  2. Choose a plan or pay as you goA monthly price with phone minutes included, or $0.004 a minute from a prepaid balance.
  3. Make a key and connectIn the dashboard, Connect AI tools gives a ready line for Claude Code, Codex, Cursor and other clients.
Get an API keySDK on GitHub
ConnectMCP · skill · Python
# Claude Code - the dashboard has the line for every agent
claude mcp add --transport http clousd \
  https://api.clousd.com/mcp \
  --header "Authorization: Bearer cl_live_…"

# the skill: how to work with phones
npx skills add https://github.com/Clousd-Android/clousd-sdk \
  --skill clousd

# or in Python
pip install clousd

Blog

Guides on this topic

All guides
AI agentsHow AI agents control an Android phone

An agent drives a phone with a loop: look at the screen, decide, act, check. How each step works, what MCP adds, and the failures every agent builder runs into.

·7 min read
DevelopersADB over the internet, safely

Four ways to stretch adb from your laptop to a phone somewhere else, the commands for each, and the habits that keep a debug bridge from becoming an open door.

·7 min read
BasicsWhat is a phone farm

A neutral look at phone farms: what they are made of, what they are legitimately used for, why ad networks and app stores treat the term with suspicion, and when cloud phones replace the rack.

·6 min read

Questions about phones for AI agents

How fast is a look and an act?

An observe without the UI tree takes well under a second; with the full tree, a few seconds. An act returns as soon as the screen settles, or after at most 5 seconds.

Which agent frameworks does it work with?

Claude Code, Codex, Cursor, Gemini CLI, VS Code and any other MCP client. Without MCP it is plain REST and JSON, and in Python there is pip install clousd.

How do I get access?

Sign up, choose a plan or add funds, and create an API key in the dashboard. There is no waiting list.