[mathcuber]

I’m curious how you're handling rate-limit backpressure at the MCP layer when an LLM blasts multiple tool calls at once.