Cursor Cloud Agents: Autonomous Dev Environments in Isolated VMs
Article··Tool
The official documentation outlines Cursor's Cloud Agents, formerly known as Background Agents. Rather than running locally, these agents execute in isolated cloud virtual machines provisioned with full development setups, including cloned repositories, dependencies, secrets, and network access. They support parallel runs, automated testing, and interactive remote desktop control.
Key points
1Cloud Agents operate inside isolated cloud VMs and do not rely on the user's local machine staying connected.
2Support for parallel runs and multi-repo workflows across separate frontend, backend, or shared library repositories.
3Triggers available across multiple channels: Cursor Desktop, Web, iOS, Android PWA, Slack, Linear, API, and PR comments on GitHub or Bitbucket.
4Deep integration with the Model Context Protocol (MCP) via HTTP/stdio and custom command hooks configured via .cursor/hooks.json.
6Billed at model API pricing with mandatory spending limits on paid Cursor subscriptions.
Why it matters
Local coding agents often struggle with isolation, parallel task execution, and reproducing full runtime environments. Providing agents with dedicated cloud VMs equipped with testing pipelines and browser control enables them to independently build, verify, and document their changes without disrupting local developer setups.
Analysis
Cursor is transitioning AI agents from code-completion tools into asynchronous cloud workers. By equipping agents with dedicated environments configured via snapshots or Dockerfiles, they can close the execution loop through testing and debugging. The combination of remote desktop access, verification artifacts, and granular team permission checks addresses both observability and trust requirements in automated development.
Cursor Automations run cloud agents in the background, triggered either on a schedule or via events from GitHub, GitLab, Bitbucket Cloud, Slack, Linear, Sentry, and PagerDuty.
Cursor highlights its expanded AI coding platform, introducing parallel autonomous cloud agents, background automations, and support for multiple state-of-the-art models across desktop and CLI interfaces.
Cursor has documented its plugin system, which bundles rules, skills, agents, commands, hooks, and MCP servers into distributable packages. Alongside its proprietary format, Cursor now supports the open Agent Plugins specification.
Cursor's official documentation details Plan Mode, an agent workflow that researches repositories, asks clarifying questions, and drafts an editable implementation plan before generating any code.
The cursor/plugins GitHub repository houses official plugins and specifications for the Cursor AI code editor, covering agent workflows, code reviews, task orchestration, and SaaS integrations.
Cursor has rolled out an early beta of Origin, its own code hosting solution. It includes Git repositories, pull requests, bidirectional GitHub synchronisation, and native integration with AI agents and deployment tools.