Run multiple Cursor sessions in parallel
Use Conductor workspaces to run several Cursor tasks at once with separate branches, files, terminals, and review state
Use this guide when you want several Cursor tasks moving at the same time without one task taking over another task's checkout.
In Conductor, the unit of independent Cursor work is a workspace. Create one workspace for each task that should have its own branch, working tree, setup context, terminal, diff, and pull request path.
Conductor supports Cursor's Composer 2.5 for Cursor sessions, so Composer 2.5 tasks can run in the same workspace model as Claude Code and Codex tasks.
Cursor also has native IDE and background-agent workflows. Use Conductor when you want parallel Cursor sessions plus Conductor's workspace sidebar, setup scripts, run scripts, diff review, checks, and pull request flow.
For the decision model, see Parallel agents. For workflow-level differences, see Compare Conductor vs Cursor.
Before you start
Cursor sessions, including Composer 2.5, need a Cursor API key. Add it in Settings -> Providers -> Cursor, or set CURSOR_API_KEY in Conductor's environment settings.
You also need a repository that can run from a workspace directory. Use Files to copy for static gitignored files such as .env.local. Use a setup script for commands that install dependencies, generate files, create symlinks, initialize local databases, or prepare other per-workspace resources.
Choose the unit of parallel work
Before you start more sessions, decide whether the work needs isolation or shared state.
Use separate workspaces when each Cursor task should become its own branch, review, or pull request.
Use multiple chats in one workspace when the tasks need to share the same branch and current code state, such as one Cursor session implementing while another reviews the same diff.
This choice matters more than the number of sessions. Parallel agents help when the tasks are independent enough to review separately.
Run several Cursor tasks
Create one workspace per independent Cursor task
Create a workspace with Command Shift N or the ... button next to New workspace.
Start from a GitHub issue, Linear issue, pull request, branch, or a new task. Conductor creates a separate git worktree and branch, then attaches the Cursor chat, terminal, diff, and review state to that workspace.
Start Cursor in each workspace
Open the workspace and choose Cursor for the chat. Use Composer 2.5 when you want Cursor's Composer model in a Conductor workspace.
Give each Cursor session a scoped prompt: what to change, what not to change, how to verify it, and when to stop for review. Repeat this for each independent task.
Run and test each workspace separately
Use the Run button or terminal from the workspace that owns the change. Setup and run scripts execute from the workspace directory.
If several workspaces need local servers at once, use CONDUCTOR_PORT so each workspace gets a separate port range. If your project cannot run cleanly from workspace directories, use Spotlight testing.
Open the workspace in Cursor when useful
If you want Cursor's editor surface for manual editing, click Open In or press Command O and choose Cursor.
Opening the workspace in Cursor does not replace the Conductor workspace. Conductor still owns the branch, terminal, diff, checks, pull request, and archive flow.
Review and merge workspace by workspace
Open the Diff Viewer with Command Shift D and inspect one workspace at a time.
Use the Checks tab to track git status, CI, deployments, comments, and todos. Merge and archive the workspaces that are ready; keep revising or archive the ones that are not worth shipping.
Keep the sessions coordinated
- Give each Cursor session a task that can finish independently.
- Avoid assigning two workspaces the same file-heavy refactor unless you expect merge conflicts.
- Keep durable project context in
AGENTS.md,CLAUDE.md, repository instructions, or the workspace.contextdirectory. - Use one workspace when the Cursor sessions need to collaborate on the same in-progress branch.
Caveats
For the same workflow with Anthropic's agent, see Run multiple Claude Code sessions in parallel. For the same workflow with OpenAI's agent, see Run multiple Codex sessions in parallel.