Same issue here. We need Cloud Agents from Slack to create documents in Outline via MCP.
Is there a way to configure MCP for Cloud Agents at team level? Or any workaround? Our use case: sales team creates offers via @Cursor in Slack — we need direct publication to Outline, not just local file.
Same, I’m a bit shocked that cloud agents have been around this long and don’t have MCP. Codegen (already acquired), EngineLabs, what not already supported it
Please let us know when this is live! I’m sure you have hundreds of user looking to control their codebase via Slack and without DB/MCP access cloud agents are largely futile cause they lack data context
Does the Feb 24, 2026 release now enable Cloud Agents to use custom MCP servers? Please explain where in the Cursor Cloud Agent API an MCP configuration can be supplied in the launch payload. I don’t see a way to configure a custom MCP server for Cloud Agents via the Cursor Dashboard.
Per Cloud Agents API | Cursor Docs , it says “MCP (Model Context Protocol) is not yet supported by the Cloud Agents API.” When will it be supported because Capabilities | Cursor Docs implies that custom MCP servers are now supported.
You can find MCP Servers for Cloud Agents now on http://cursor.com/agents A bigger launch is coming this week, but we’d love for you to try it out already!
Cloud Agents created via the API should still be able to call MCPs, as long as you’ve enabled the MCPs for that repo via the Agents dashboard.
Colin, will Custom MCP servers be configurable programmatically on an Cloud Agent-by-Cloud Agent basis? Meaning is there an API endpoint to launch a Cloud Agent with a custom MCP server attached to it rather than manually having to manually define Custom MCP Server details at the Dashboard level?
Thanks! I was able to configure Linear and Supabase. A bit confused as to how these work per repo. Am I able to configure MCPs per repository ? When I initiate a cloud agent from slack for example, do I tell it to use certain MCPs or will they always have access ?
And if I create the mcp at cursor.com/agents, there is no way that I have been able to figure out for the smartbear one to pick up those env secrets. If I use the cloud agent terminal, I can echo them, so they’re on the VM, but the MCP has no ability to see them. I’ve tried with no env block in the mcp config, the ${env:} interpolation thing, and putting the actual key / value secrets into the custom MCP setup at cursor.com/agents. In every case, the SmartBear tool says it can’t find the auth token and API key. Works fine locally in Cursor.
Please help! Cloud Agents are relatively useless to me if you can’t interact with any external services except the handful in the marketplace…
@ryanw I can confirm that configuration interpolation isn’t working for MCP Servers configured for Cloud Agents. When configuring the MCP, you’ll need to use the actual secret value.
I set up a SmartBear account to verify, and everything is working for me with this configuration:
Can you try reconfiguring your SmartBear MCP in the portal with the actual secret values (which are still encrypted in the database) hardcoded in the env field (not using ${env:...} interpolation), and let me know if that works for you?
I’ll file a report with the team about the config interpolation!
Thanks for the update. Configuring via the dashboard works.
Can you please consider establishing a single source of truth for MCP config across local and agents? Having duplicated definitions across two mcp.json files means they may well drift apart over time, and it adds friction to the “it just works” process of firing up a Cloud agent.
Unfortunately, it’s still not working for me, even if I edit the mcp json (on cursor.com/agents) and put the token and key directly into the json. Agent always says the same thing:
" BugSnag MCP tools are available but not configured. Both calls returned:
“The tool is not configured - configuration options for BugSnag are missing or invalid.”
I’ve double-checked that my keys and values are the same in local (where this works fine) as cloud…could this possibly be related to whitelisting domains?