Cursor Multi-root Workspaces: Cross-Repo Agent Sessions

Multi-root Workspaces: One Agent, Many Repositories

Cursor 3.2 introduces multi-root workspace support in the Agents Window — a capability that extends a single agent session beyond the boundaries of a single repository. In a multi-root workspace, multiple project folders are opened together as one unified workspace, and Cursor's agents can operate across all of them simultaneously.

Cross-Repository Changes Without Context Switching

The most significant impact of multi-root workspace support is the elimination of manual retargeting. Previously, when a code change needed to propagate across a frontend application and a backend API — for example, updating an API contract that requires changes in both — a developer either had to maintain separate agent sessions in different IDE windows or manually re-point the agent at each repository in turn.

With multi-root workspaces, a single Cursor agent can make coordinated changes across a frontend, backend, and shared library in one session. As Cursor describes it, the agent can "make cross-repo changes spanning frontend, backend, and shared libraries, without retargeting the agent every time it moves between repos."

Project Hooks from All Workspace Folders

A practical improvement bundled with multi-root workspace support is that Cursor now reads project hook files from all folders in the workspace, not just the first one. Hook files are project-specific configuration files that define how agents should behave in a given repository — coding conventions, lint rules, context instructions, and similar settings. By reading hooks from all workspace folders, Cursor ensures that agents behave correctly in each repo's context, even when transitioning between them mid-task.

Who Benefits Most

Multi-root workspace support is particularly valuable for teams that maintain monorepo-adjacent architectures: separate repositories for different service layers (API, web, mobile, shared packages) that still need to evolve together. Full-stack developers, platform engineers, and anyone who routinely makes changes that span repo boundaries will find that multi-root workspaces remove a friction point that previously required either complex tooling or significant manual coordination.

This feature also pairs naturally with the /multitask command — while one set of subagents works on the frontend portion of a cross-repo change, another can simultaneously address the backend, all within the same unified workspace session.