Fingers crossed… hope you get a useful response.
My support case currently seems stalled - no response yesterday after my last update.
In case any OAI person sees this thread, this is one of the errors I see when loading a workflow. Sure seems like loading something in the workflow config throws an exception and it just dies. And looking at this snippet, sure does look like MCP headers.
Unhandled Promise Rejection: TypeError: undefined is not an object (evaluating ‘B.expression.replace’)
I think openai should allow us to get rid of the MCP nodes affecting the workflows.
Seems like a few things:
- If the current revision cannot be loaded, the builder should offer to load the last working one or at least make it accessible. Right now it just errors out.
- If a revision cannot be loaded, they could offer some sort of “clean up” tool and maybe deactivate problematic nodes.
- Or really, they should just fix the MCP nodes so that they don’t blow up when you add headers.
Got a note over the weekend that it was being looked at by an engineering team. Anyone else heard anything else that’s new? Love to keep the momentum up on this until it’s fixed. @OpenAI_Support
Hi everyone! We deployed a fix for the issue where adding an MCP task and selecting “Add Server” or “Add Connection” caused the “Add connector” modal to freeze with an indefinite loading spinner and we're no longer able to repro the issue on our end.
Are you all still seeing the “Failed to load workflow” error or other errors?
2 Likes
@OpenAI_Support That sounds like progress, thanks.
Yes, can confirm that any of my workflows that contained MCP servers with header based authentication no longer load and show “Failed to load workflow”.
3 Likes
@OpenAI_Support I just started to get the same error today - it was working fine yesterday and no changes were done tot he workflow since. The workflow does contain an MCP Server with header based authentication
@OpenAI_Support I have been getting the same error, i tried accessing a older version and changing the MCP authorisation, but has not solved anything. Is there a plan to deploy a solution to this issue?
@OpenAI_Support I have the same issue. For more information my workflow contained Pipedream’s Discord MCP connector.
@OpenAI_Support Anything new on this?
No confirmation from OAI but this appears to be fixed? All of my broken workflows now load without error. 
2 Likes