Hi there,
I have build an AI Agent that was working like a charm for more than 3 weeks. All a sudden, it doesn’t work anymore and I got this error message : “Could not connect to your MCP server”.
I’m struggling since two days trying to fix this issue. Without success. Even after reading this other related topics :
I also tried :
- changing the URL of the MCP including localhost:5678
- same thing but with the IP of the server
My set up :
- n8n version : 1.102.4
- Agent IA node version : 1.9
- MCP node version : 1
- MCP Trigger version : 2
- Hébergement : VPN Hostinger
You also find below a test workflow with the same issue :
The strangest thing is that it was actually working for weeks, and all a sudden it doesn’t work anymore…
Does anyone already faced this issue ? Any idea on how to fix it ?
Thanks a lot !
Have you tried to … de-activate and re-activate the workflow, which hosts the MCP Server trigger?
Another things, which slightly concerns me is that the url to your server
https://n8n.srv842036.hstgr.cloud/mc-testp/40b88f43-43bd-48c5-98aa-c71554475a60
doesn’t have /mcp/ in it, but does have /mc-testp/… Could you check on that URL? Try re-creating the trigger from scratch as well.
Mine looks like this:
/mcp-test/<uuid-path> for the test URL and
/mcp/<uuid-path> for the production URL
1 Like
Hey @jabbson, thanks a lot for your help !
So I’ve tried to de-activate/re-activate the workflow. I also tried to create a new workflow. Desactivate node by node… but the issue is still here.
You’re right for mcp-testp ; I’ve corrected the path. Still not the solution 
Ok, let’s continue our discussion here!
You said that you’re using hostinger, BUT… your MCP Server Trigger is still in localhost (as seen in below)
http://localhost:5678/mcp-test/40b88f43-43bd-48c5-98aa-c71554475a60
Can you confirm this?
Yes, i’m going to delete the other thread ; you’re right.
You’re right. My MCP Server Trigger is localhost:5678 as I’ve read on the community that it could solve my issue. However, even if my usual url (like https://n8n.srv844536.hstgr.cloud…) it doesn’t work.
While I was writing to you, I’ve found something even more confusing.
This Agent + MCP is working well now (it didn’t yesterday) :
While this one, which is the one I’m trying to debug and have the pretty same configuration, isn’t working :
I’m so confused !
Do the MCP Client tools is executed when you’re testing it?
Yes, the MCP Client tools is executed, but MCP Server isn’t triggered :
Can you open the MCP Client node when the error is triggered?
Sure, here what I see when opening the MCP Client Node after the error is triggered :
Kinda confusing tho…
What if you separate the mcp client and mcp server workflow?
So, you will have two workflows
Sure @jabbson,
Here’s a screenshot from the config screen of the MCP Server
I’ve started by designing two separates workflows, and it was working well (last week). Since it doesn’t work anymore, I’ve assembly the two workflows in the same workflow. I’ve kept the other MCP server workflow separately. I’ve tried both of the path (the one that is calling the separated MCP Server, and the one that is calling the integrated MCP Server). In both case, same issue.
In both case, it doesn’t work @jabbson, @cutecatcode
Interestingly when I try to access your server, I can:
In your client I do not see authorization used, do you use any auth to protect the server?
When I try to access it via mcp client, I am prompted to enter credentials, so there must be some configured
If you’re using mcp-test, try to change with /mcp/ instead of mcp-test.
Try to run it again with separating the MCP workflow.
damn this is so strange ! I have no idea what to do..
For the moment, I don’t use any protection (I know I should 
)
Thanks for the suggestion @cutecatcode !
When running with /mcp/ instead of mcp-test on the same workflow
When running with /mcp-test/ in test mode on the same workflow
When running with /mcp/ in a separate workflow
When running with /mcp-test/ in the separate workflow
I’m so lost…
While you’re running with /mcp/, do you activate the MCP Server workflow?
Yes, the MCP Server workflow is active.
I’ve also tried to click on “execute workflow” in MCP Server workflow, before launching the other workflow : same issue
Kinda weird tho…
I think it’s server issues (?)
I tried with @jabbson test flow, and have same results
1 Like
Hi @RomainMaltrud, have you ever been disabling the gzip conversion (if it’s available in hostinger)?
