Skip to content

Cloud Agents

Cloud Agents use the same agent fundamentals but run in isolated VMs with a full development environment: cloned repos, installed dependencies, secrets, startup commands, and network access.

Official: Cloud Agents. Former name: Background Agents — one product, not two.

This page is the form: when to use it, how to start it, what differs from local. Task recipes stay in the cookbook.

Prerequisites

  • A paid Cursor plan
  • An account admin has connected source control: GitHub, GitLab, Bitbucket Cloud, or Azure DevOps
  • Read-write access to the repo (and any dependent repos or submodules)
  • You accept that Cloud is officially the feature that requires Cursor to store code. If policy forbids storage, leave Cloud off

Learning objectives

After this page you can:

  1. Decide Cloud vs local Agent vs Bugbot vs CLI
  2. Start a run from desktop, web, mobile / PWA, Slack, GitHub, Linear, or agent + &
  3. Treat environment setup as the main lever
  4. Know the official MCP / Hooks limits on the VM

Why use Cloud Agents

  • Run as many agents as you want in parallel
  • Your laptop does not need to stay online
  • The VM can build, test, and interact with the changed software, including desktop / browser computer use
  • Team MCP servers (HTTP and stdio; OAuth when the server needs it)
  • Multi-repo environments: inspect the workspace, make coordinated changes, open a PR per repo it changes. Official: long-running is not available for multi-repo yet

Official wording: skipping environment setup is like not giving engineers a computer.

How to start one

SurfaceHow
Cursor DesktopAgent input → Cloud in the dropdown
Cursor Webcursor.com/agents on any device
Cursor for iOSNative app — mobile
AndroidChrome → cursor.com/agentsInstall App (PWA)
Slack / Linear@cursor (admin installs the app first)
GitHub / BitbucketComment @cursor on a PR or (GitHub) issue
APICloud Agents API
Cursor CLIPrefix a message with & — see CLI

Do not send “three lines under the caret” here — that is Tab. Do not use Cloud as a PR reviewer — that is Bugbot.

How it works

  1. Cloud clones from the connected provider onto a separate branch
  2. The agent works in the configured environment
  3. It pushes and hands off (often a merge-ready PR plus artifacts)

Configure the environment with agent-led setup, a saved snapshot, or a Dockerfile via .cursor/environment.json. See Cloud agent setup. Builds prepare repos and dependencies in the background.

Hover the repository name on an agent page to see which environment and Build that run used.

Runtime controls (official): secrets, outbound domain allowlists, Tailscale / similar for private networks, private connectivity for supported SCM paths. Full set: Security and network.

MCP and Hooks (official limits)

MCP: team servers from the MCP dropdown on cursor.com/agents. HTTP and stdio. OAuth supported. Built-in Cursor Cloud MCP for run diagnostics (transcripts, events, environment, setup logs).

Hooks: command-based hooks from .cursor/hooks.json in the repo. Enterprise also runs team hooks and enterprise-managed hooks. They do not run during early read-only exploratory turns; they start once the agent has a writable environment.

Not available on Cloud:

  • User-level ~/.cursor/hooks.json (the VM has no access to your home directory)
  • IDE-specific hooks (Tab hooks, workspaceOpen)

Supported families include tool / file hooks (preToolUse, beforeShellExecution, afterFileEdit) and lifecycle hooks (beforeSubmitPrompt, subagentStart / subagentStop, preCompact, afterAgentResponse / afterAgentThought, stop). Matrix: Hooks · Cloud agent support.

Artifacts, desktop, and sharing

  • Artifacts: screenshots, videos, logs — what changed and how the agent verified it
  • Remote desktop: take control of the agent's desktop to test without checking out the branch; release control so it continues
  • Share: send the agent URL. Viewers must be on the same Cursor team and must connect their own SCM account with access to that repo. Team membership alone is not enough. Viewing is read-only unless a team admin enables team follow-ups

Models and billing

Cloud Agents use a curated model list. You can select context-window size on supported models. Larger windows can increase tokens and cost.

Charged at API pricing for the selected model. You set a spend limit the first time you use them.

Mobile, JetBrains, Plugins, Automations

Index-level only — official pages, not extra tutorials:

SurfaceOfficial one-liner
iOS / PWANative iOS app (beta, iOS / iPadOS 26+) starts and reviews Cloud Agents. Android: install the PWA. Mobile
JetBrainsACP agent inside IntelliJ / PyCharm / WebStorm. Paid plan + AI Assistant 2025.1+. JetBrains
XcodeXcode 26.3+ built-in MCP: build, test, SwiftUI previews, Apple docs. Xcode
PluginsBundles of rules, skills, agents, commands, MCP, hooks. Plugins
AutomationsCloud Agents on a schedule or events (GitHub, GitLab, Slack, webhooks, Linear, …). Bugbot, Security Agents, and PR Routing are the three Cursor-managed agents on that page. Automations

When to use it

CloudStay local
Away, overnight, many parallel runsYou must watch the diff, edit a plan, or use Debug Mode
Isolated VM + tests + PRSecrets / services that never left your machine
Coordinated multi-repo PRsOne repo you already have checked out

Task recipes (environment.json, AGENTS.md heading, dispatch checklist): Cookbook · Cloud.

Common pitfalls

PitfallDo this
Skip environment setupAgent-led setup or .cursor/environment.json + Secrets tab
Expect local mcp.json / ~/.cursor/hooks.jsonTeam MCP on cursor.com/agents; project command hooks only
Use Cloud as the PR reviewerBugbot
Spin a VM for three linesTab or Cmd+K
Multi-repo long-runningOfficial: not available yet
Teammate cannot open your runSame team and their own SCM access to that repo
Put .env.local in a snapshot and call it doneOfficial preference: Secrets tab
Document “Background Agents” as a second productSame product; new name is Cloud Agents

Next steps

Built for frontend engineers · Powered by VitePress