Allow for agents to communicate with each other

Ideas Feature Requests
1

Feature request for product/service

Chat

Describe the request

Hi Cursor team,

I’d like to propose a feature that would significantly improve multi-agent workflows: within one agent instance, allow asking a question to another agent instance, where the other agent returns content that is automatically inserted into the current agent’s context.

The core value is specialization. For example, I could maintain a frontend-focused agent that has deep knowledge of frontend architecture, conventions, and relevant files, and a backend-focused agent for APIs/services. Then, instead of having the backend agent repeatedly read large portions of the frontend to answer cross-cutting questions, it could query the frontend agent and immediately use the returned information.

This would also help Cursor scale to much larger repositories: instead of forcing a single agent to load a huge codebase into context (or re-load it repeatedly), knowledge can be partitioned across specialized agents and pulled in on demand—effectively making far larger projects workable within context limits.

Thanks for considering—I’m happy to share concrete examples from real projects, as well as the specific workflow I have in mind, if that would be helpful.

Best,
Philipp

Operating System (if it applies)

MacOS

2

For now, theoretically, you could ask the Cursor Agent to launch Gemini CLI or similar at some point during its runtime in its terminal.

Although it would be interesting to implement the entire Manus AI inside a single Cursor window.

3

Hey, thanks for the request.

Multi-agent collaboration and orchestration is something users have asked for before, and the team is aware of the interest in this area.

For now, there’s a community solution worth checking out. You can run and orchestrate cursor-agent instances via the CLI directly from inside Cursor. See this recent post for details: How to Use Cursor-Agent from Inside Cursor IDE This might help some folks with complex workflows: How to Use Cursor-Agent (and/or Claude-Code-SDK) from Inside Cursor IDE.

I’ll pass your specific use case to the team, like separating frontend and backend work for large repos.

4

1 year on and this thread is still the main place this concern lives. Just posted a more specific shape for how it might be implemented — framing it as “we’ve built the mailbox (shared folder), we just need the doorbell (a one-line chat-notify primitive that injects a turn while preserving context).” Details and field-report: Feature request: chat-notify primitive — we already have the mailbox (files), we just need the doorbell

Curious if the shape there matches what you (OP) actually want, or if you’d expect something different.