Hey everyone,
I’ve been diving into custom agents in Cursor lately, and it really feels like we’re hitting a turning point in how we interact with our projects. With the arrival of multi-chat, we’re clearly moving toward a smart multi-agent setup, where each agent could have a specific role.
So, a few questions for those of you who’ve already explored or optimized your setup:
• What are your favorite agent combos? Do you assign them specific roles or keep things more general-purpose?
• Do you use one agent per task (refactoring, unit test generation, clean code, debugging, etc.)? If so, how do you split the responsibilities?
• What kind of context do you provide to your agents? (Currently limited to text only: prompts, custom instructions, internal guidelines, etc.)
• Are your agents using MCPs? Which ones do you consider essential or particularly effective?
• Do you combine Cursor with other AIs (ChatGPT, Claude, Perplexity…) in your workflow? If so, why, and how do you divide the tasks?
I really believe Cursor’s full potential will come to life with a well-thought-out multi-agent approach.
Looking forward to hearing your setups, your experiences, and your best practices!