How to use MCPs with self-hosted agents?

Support Help
, ,
1

Hello folks!! The documentation in the self-hosted page is too concise and I’m not entirely sure on how to setup our internal mcps. Some questions that I have are

  • Does self-hosted agents mcps use only team or only repo definitions?
  • How do the self-hosted agents load the mcps and how can i debug it?
  • For our internal tools in a vpc, is it required to use stdio type?
  • Where the cloud agents dashboard is for me to configure this mcps? I see the team mcps options

Thank you in advance!!

image
image1075×594 120 KB

image
image1071×412 51.8 KB

3

Hey, let me go through your questions:

Team vs repo MCP?
Cloud agents only pick up MCP servers configured through the dashboard at https://cursor.com/dashboard/integrations. Repo-level mcp.json isn’t read by the cloud agent runtime.

Stdio for VPC tools?
Yep. If your MCP server needs to reach private networks or internal APIs behind your firewall, use command stdio transport. HTTP and SSE servers are routed through Cursor’s backend, so they won’t reach private endpoints. More details here Self-Hosted Pool | Cursor Docs.

Dashboard location?
Use https://cursor.com/dashboard/integrations for team MCP config. When you start a session at cursor.com/agents, you’ll see toggles to enable them per run.

Important note: self-hosted pool requires a Team plan. On the Free tier, this feature isn’t available.

For debugging MCP loading on self-hosted workers, there’s no dedicated tooling for that right now, unfortunately.

Let me know if anything else comes up.