I notice in the otel traces that call_tool often has ~300-500ms of latency before the the execute / mcp-graphql-client call happens (this is for tools defined in operation files).
Hi @glenn that is much higher than we have seen in our testing, thank you for reporting! Can you give us some more details about how you are running the server and how many tools you have registered? Are you seeing the 300+ms consistently on every request or only on the first tool call?
We have 5 tools via operation files. We consistently see the overhead, not just on the first tool call. The built-in tools seem to be more stable with around 30ms overhead (which is still pretty high I think).