Comment by the_arun

In a multi model situation, shouldn't LLM A talk to LLM B as a tool call via MCP? or would it talk to LLM B directly?