Are you using our SaaS platform (Baserow.io) or self-hosting Baserow?
Self-hosted
If you are self-hosting, what version of Baserow are you running?
1.33.4
What are the exact steps to reproduce this issue?
Hi Baserow community,
I’m experiencing an issue with MCP (Model Context Protocol) endpoints when deploying Baserow on Amazon EKS, and I’m hoping someone can help me understand what I might be missing.
My Setup:
Baserow version: 1.33.3
Deployment: Amazon EKS using the official Baserow Helm chart (v1.0.24)
Issue: All /mcp/* endpoints return 404 errors
What Works:
I have an identical Baserow deployment on Amazon ECS that works perfectly
The ECS deployment uses a single all-in-one Baserow image and successfully serves MCP endpoints like /mcp/{token}/sse
Both deployments use the same Baserow image version (1.33.3)
What Doesn’t Work:
On EKS, any request to /mcp/* paths returns: {“error”: “URL_NOT_FOUND”, “detail”: “URL /mcp/{token}/sse not found.”}
I’ve confirmed the ingress routing is working (traffic reaches the backend)
The Baserow application itself works fine - it’s only the MCP endpoints that are missing
Hey @franciscocunha57, I’m sorry to hear that you’re having problems with the MCP server in your cluster. If the endpoints respond with a 404, then it looks like either something in the routing is not working as expected, or the Baserow version of one of the pods is still on an older version.
If you visit: https://api.YOUR_DOMAIN/api/redoc/, which version do you then see between the brackets of the title?
I’m also curious if the ASGI pods are on the right version. If you run the following command on your cluster kubectl -n YOUR_NAMESPACE describe pods | grep -E 'Image:|Name:', how does the output then look like?
Hey @franciscocunha57, your definitely running the right version of Baserow. In order for the MCP server to work, the HTTP traffic must be routed to the ASGI workers (same as the webhoks). All other HTTP requests are routed to the WSGI workers.
