Agent ACP sometimes fails with "We're having trouble finding the resoure you requested"

Support Bug Reports
,
1

Where does the bug appear (feature/product)?

Cursor CLI

Describe the Bug

When using Cursor CLI as an ACP server (cursor-agent acp), I frequently see this error returned:

{"jsonrpc":"2.0","method":"session/update","params":{"sessionId":"…","update":{"sessionUpdate":"agent_message_chunk","content":{"type":"text","text":"\n\nError: I: Provider Error We're having trouble finding the resource you requested. If the problem persists, please contact support."}}}}

Sometimes prompting the agent to continue or restarting the ACP server resolves the issue.

My other workaround is to run cursor-agent --resume and finish the in-progress work that way. I don’t seem to encounter any such problem when running cursor-agent directly.

I’m not sure how to investigate further.

Steps to Reproduce

I don’t have reliable repro steps.

I’m using agent-shell in Emacs to connect to cursor-agent acp. Sometimes the error shows up right away, other times it shows up after the session has been active for a while.

Expected Behavior

I would at least expect a more detailed error response to be able to understand the issue better. Maybe there’s a debug option that could be added to cursor-agent acp or something?

Operating System

MacOS

Version Information

% cursor-agent --version
2026.03.20-44cb435

Does this stop you from using Cursor

No - Cursor works, but with this issue

5

Hi @Stuart_Campbell_Via

Thanks for the detailed report.

This error (“We’re having trouble finding the resource you requested”) occurs when an upstream model provider temporarily returns a 404 for the Auto-selected model deployment. It’s transient and typically resolves on retry.

You’re right that ACP surfaces this more than direct CLI mode. There is a difference in error recovery behavior between the two: direct CLI automatically retries certain transient provider errors, while ACP does not currently have the same retry logic. We’ve noted this as an improvement for ACP.

Workarounds:

  • Retry the prompt – Auto mode will often route to a different provider on the next attempt

  • Pin a specific model (e.g., claude-sonnet-4) to avoid intermittent routing issues

If it keeps happening, you can grab the request ID via /copy-request-id in your ACP session so we can trace the specific failure.

1 Like