MCP routing issue in prod only

[jamesamcl]

In wwwdev we have one backend instance, MCP works fine

In www we have two backend instances. Sometimes MCP SSE messages go to one of them and sometimes the other. But the session state is not shared so this causes errors.

Possible solutions:

1. Share session state across the backend instances somehow
2. Route all MCP requests to one backend
3. Scale down to one backend instance (with more resources?)
4. Use streamable HTTP MCP instead which may not have this problem

For now I have updated the recommended MCP URLs to wwwdev