Been running Cursor Cloud Agents for a couple weeks now and wanted to share a setup that finally made them useful for real end-to-end dev work, not just isolated code changes.
The problem: Cloud agents are great environments, but out of the box they’re cut off from the two things I actually needed
- access to my project databases
- the ability to SSH into my servers
What I built:
Private MCP layer
I set up an MCP server on a VPS that sits between my agents and my projects’ databases (also added Email access - gmail, imap & resend). It has its own dashboard and DB, so it’s not tied to my laptop at all. I issue scoped API keys per tool. Built it for Cursor Cloud originally, but since most agentic tools support custom MCPs now, I can plug any agent into the same layer.
SSH access for agents
I created a dedicated cursor-agent SSH user and new key pairs for my staging (and some lower-risk prod) boxes. The private key gets passed in as an env var to the cloud environment, and a short startup script wires up SSH access automatically. Now agents can read logs, restart services, and actually debug things;
Result: for my workflow this was the unlock to go fully cloud agentic and now most of my chats are default on cloud. Then I just let them work and control them with my phone and my PC does not have to be on all the time
Anyone has built anything similiar or has any tips to better this setup?