Cursor just turned your phone into a coding agent dispatcher

Cursor, the AI-powered IDE, released a broad update on Tuesday that reworks how developers interact with its coding agent, adding features that reduce context switching and extend the assistant's reach beyond the editor.

The most visible change is parallel chats. Developers can now open secondary conversations alongside their main agent chat using the /side or /btw commands, or a plus button in the chat panel. Each parallel chat is a full, persistent conversation thread. You can revisit it later, or inject its context back into the main thread with an @ mention. By default, these side chats are read-focused: they handle clarifications, alternative explorations, and quick verification without derailing the primary agent's execution.

Cursor also added full-text search across agent transcripts. In the Agents window, developers can search through thousands of past conversations from the command palette using Cmd+K. Cursor builds a local search index for speed. Within a single conversation, Cmd+F provides in-document search with result counts and navigation.

Slack gets multi-repo teeth

Cursor's Slack integration, introduced earlier, now carries more weight. The agent shares its plan before starting work and updates its status as it progresses, so team members can intervene early. It can operate inside named multi-repo environments. If a front-end, back-end, and shared code live in separate repositories, Cursor picks the right environment based on the request. During a task, if it needs a repo outside the current environment, a simple "Switch Repository" button lets the user select one, and the agent resumes where it left off.

The assistant can now also read and send messages in other Slack channels and threads, pulling context from elsewhere in the workspace and posting updates back to the relevant thread or channel. The message formatting has been cleaned up: inline buttons are gone, replaced by compact links in the footer, and tables, PRs, and artifacts display more cleanly.

Mobile agents and remote control

Cursor launched a mobile app that lets users pick a repository and launch an agent from their phone, using voice input with any leading model. Cloud agents run in isolated virtual machines with full development environments. Sessions can be moved from local to cloud so they keep running when the laptop is closed. Remote Control, available on Teams and Enterprise plans with admin activation, lets a user grab an agent session that started on their computer and continue steering it from the phone. Live Activities on the lock screen show agent status, and push notifications alert when an agent finishes, needs input, or is ready for review. Developers can view demos, screenshots, logs, and diffs from the phone, add follow-up instructions, or merge a PR on the spot. This brings the IDE's notorious agent-first philosophy to your pocket, a shift that began with Cursor 2.0's agent-first redesign.

Customization hub and team MCPs

A new Customize page centralizes plugins, skills, MCPs, sub-agents, rules, commands, and hooks, all configurable at user, team, or workspace level. Plugins now support ready-made canvases: shared templates teams can open and reuse, such as a Hex Canvas for data visualizations or an Atlassian Canvas for real-time issue and project views.

Administrators can configure team MCP servers once and deploy them across Cloud Agents, the Agents window, the IDE, and the CLI. Those same servers can appear in a team marketplace under Dashboard > Integrations & MCP, letting team members install approved integrations locally without setting up servers themselves. The marketplace now supports organization groups alongside SCIM directory groups, giving admins finer control over access. As other model providers race to make agent deployments simpler, Cursor's approach echoes a broader trend in one-click cloud agent setups.

Team marketplaces also now support importing plugin repositories from GitLab, Bitbucket, or Azure DevOps.

Refined project and repository selectors

The update streamlines project and repository pickers. Branch selectors open to the default branch and recently used branches instead of a long linear list. The Home concept has been removed. Working without a repo is now an explicit No Repo choice. Multi-repo and multi-root selection collapses into a Select Multiple button within cloud and local panels. All remote options (personal machines, team pools, existing remote workspaces) sit under a single Remote Machines menu with search. Repository search is now scoped to where you're working (local, cloud, or a specific remote machine) rather than being a single global search. This level of granularity in agent orchestration is precisely what recent research into agent cost and observability warns is needed for production use.