Cloud Agents
Cloud Agents run in isolated cloud environments instead of on your local machine.
What can Cloud Agents do?
Cloud Agents handle coding tasks without you needing to be in the loop. They can build features, fix bugs, write tests, open PRs, and share a video of it working. Each task starts from a configured cloud environment with the repos, dependencies, secrets, and network access the agent needs.
Cloud Agents can also use multi-repo environments. This lets one agent work across frontend, backend, infrastructure, or shared-library repos in the same run.
How does "Move to Cloud" handle my file state?
"Move to Cloud" does not snapshot your local uncommitted changes. The cloud agent starts from a clean git state on the remote repository. It transfers your conversation history and context, but not dirty files or uncommitted edits. Commit or stash your changes before moving a conversation to the cloud if you want the agent to work from your latest state.
Can Cloud Agents test my app in a browser?
Yes. Each cloud agent runs in its own isolated VM with a full desktop environment. Agents use a mouse and keyboard to control the desktop and browser, the same way a human developer would.
This means agents can start dev servers, open the app in a browser, click through UI flows, and verify their changes work before pushing a PR.
How do Cloud Agents show their work?
Agents attach screenshots, videos, and log references to the PR so you can validate changes without checking out the branch.
Can Cloud Agents use my personal skills?
Yes. Turn on Sync Skills for Cloud Agents under Settings → Agents to move your personal skills into synced storage for your own Cloud Agents. Synced skills stay private to you. See Use personal skills with Cloud Agents.
Project skills in the repo are always available. To share a skill with teammates, publish it to your team marketplace.
Can Cloud Agents use MCP tools?
Yes. Add and manage MCP servers through the MCP dropdown at cursor.com/agents. This gives agents access to databases, APIs, and third-party services during their runs.
Cloud agents support HTTP and stdio servers, plus OAuth for servers that need it. See the Cloud Agent capabilities page for setup details.
On Self-Hosted Machines, MCP behavior differs. See Self-Hosted Machines.
Do Cloud Agents load my team marketplace plugins?
Yes. A Cloud Agent installs your marketplace plugins when it starts and keeps those versions for the whole run. Plugins from marketplaces imported from GitLab, Bitbucket, or Azure DevOps aren't installed in Cloud Agents, and the agent starts without them. To use those plugins in Cloud Agents, import the marketplace from a GitHub repository. See Plugins.
Do Cloud Agents run hooks?
Yes. Cloud Agents run command-based hooks from .cursor/hooks.json in your repository. On Enterprise plans, they also run team hooks and enterprise-managed hooks.
Supported hooks include tool and file hooks (beforeShellExecution, afterShellExecution, beforeReadFile, afterFileEdit, preToolUse, postToolUse, postToolUseFailure) and conversation-level lifecycle hooks such as beforeSubmitPrompt, afterAgentResponse, afterAgentThought, stop, subagentStart, subagentStop, and preCompact.
Conversation-level hooks let you observe prompts, responses, and thinking; control subagents; react to compaction; and run logic when a turn completes. You can use them to build self-correcting loops around agent output and reasoning.
Hooks start once the agent has a writable environment. IDE-specific hooks like Tab hooks and workspaceOpen don't apply. sessionEnd doesn't apply on Cursor-managed Cloud Agents because they have no editor-lifetime session boundary. On Self-Hosted Machines workers, sessionStart and sessionEnd run when a session claims the worker and when that claim is released. Prompt-based hooks also aren't supported in cloud agents. User-level hooks from ~/.cursor/hooks.json aren't loaded since the VM doesn't have access to your local configuration. Hooks packaged in plugins don't run either; add them to .cursor/hooks.json in the repository or configure them as team hooks.
See the full hooks support matrix for details.
On Self-Hosted Machines, hook support matches managed Cloud Agents with a few differences. See Self-Hosted Machines.
Can Cloud Agents fix CI failures?
Yes. Cloud Agents automatically try to fix GitHub Actions failures on PRs they create. They ignore checks that are also failing on the base branch. This is currently available on Teams plans.
To disable this on a specific PR, comment @cursor autofix off. To re-enable it, comment @cursor autofix on. You can also disable it globally from Cursor Dashboard > Cloud Agents > My Settings.
How do review rules treat commits from Cloud Agents?
Cloud Agent commits are authored as the agent, not as the person who started it. See Commit author. Review rules that check the commit author or committer see the agent's identity, so the person who started the agent isn't treated as the author. To require a second reviewer on agent commits, match on the agent's author identity rather than the Co-authored-by trailer, which isn't added for every agent.
What do I need to use Cloud Agents?
Cloud Agents are available on eligible paid plans. On The X Subscription, that means Premium and above, Teams, and Enterprise; Free and Plus don't include Cloud Agents. The legacy Start, Pro, Pro+, and Ultra plans also include them. See cursor.com/pricing for the full comparison.
How do I set up Cloud Agents?
Create a new environment in your Cloud Agents dashboard. You'll connect your GitHub, GitLab, Azure DevOps, or Bitbucket Cloud account, select one or more repositories, add secrets or environment variables, and verify the setup. See the full setup guide for multi-repo environments, environment-scoped secrets, network access, and Dockerfiles.
How do I use Docker in a Cloud Agent environment that uses Builds?
A Build keeps Docker's data directory (/var/lib/docker), so images and stopped containers carry over to new agents. The Docker daemon doesn't keep running, so start it again in your environment's start command.
How do I use the redesigned project and repo picker?
The project picker keeps common setup steps in one place. You can:
- Create a project without leaving the picker
- Connect GitHub, GitLab, or Azure DevOps inline
- Search projects scoped by location: This Computer, Cloud, or a remote machine
- Remove a project from Recents with one click
Remote Machines are consolidated in the same picker flow so you can open local, cloud, and remote work from one place.
How do I start a Cloud Agent task?
- In Cursor: Select Cloud in the dropdown under the agent input
- On the web: Go to cursor.com/agents
- Via Slack: Use @Cursor
- On GitHub: Comment @cursor on a PR or issue
- On Bitbucket Cloud: Comment @cursor on a PR
- In Linear: Use @Cursor on an issue
- Via API: Use the Cloud Agent API
What are the attachment limits for Cloud Agents on the web?
At cursor.com/agents, documents and images have a 4 MB per-file limit. Videos have a separate 15 MB per-file limit. The browser rejects files over these limits before upload, so changing the model's context window or enabling Max Mode does not increase the attachment limit.
Supported document formats include PDF, Markdown (.md and .markdown), plain text (.txt), CSV, JSON, HAR, XML, YAML (.yaml and .yml), TOML, INI, and log files (.log).
If a file is too large:
- Keep only the content the agent needs, or split the file into parts of 4 MB or less
- For logs and structured text such as JSON, CSV, XML, or HAR, keep the relevant time range and requests instead of the full capture
- Start the Cloud Agent from Cursor desktop and attach the file there, since attachment limits can differ from the web app
- For non-sensitive files already in source control, ask the agent to read them from its checked-out repository
ZIP files are not supported as web document attachments. Before attaching diagnostic files, remove credentials, cookies, headers, source code, and personal data you do not want to share.
How is Cloud Agent usage priced?
Cloud Agents use the same plan usage as Agent in the editor and the CLI, at the selected model's rate. On legacy request-based plans, they still draw from your included requests, but they always run in Max Mode, so they're priced at API rates plus the Max Mode upcharge and count against your included requests as request-equivalents rather than one request each. On-demand usage takes over after the quota. See Models & Pricing for how Cursor models and third-party models bill on your plan. You can select the context window size for supported models. A larger context window can increase token usage and costs.
Can I run Cloud Agents automatically or on a cron?
Yes. Use Automations to run Cloud Agents on a schedule or from events in GitHub, Slack, Linear, PagerDuty, or webhooks. Set them up at cursor.com/automations. See the automations docs for trigger types and templates.
Does Privacy Mode work with Cloud Agents?
Yes. Cloud Agents are available with Privacy Mode. See the Cloud Agent security page for details.
Can Cloud Agents federate into AWS or other clouds without long-lived keys?
Yes. Cursor-managed Cloud Agent VMs can mint short-lived OIDC JWTs from a local socket. Verifiers use Cursor's public discovery and JWKS endpoints to check the signature and authorize on claims such as sub, team_id, and cloud_agent_id.
Can Cloud Agents read their own id, owner, or current turn from the VM?
Yes. Cursor-managed Cloud Agent VMs serve an agent metadata API on the same local socket as OIDC tokens. A GET of a path such as /v1/meta-data/agent/id returns the bare value as text.