It does not. Context is context no matter how you process it. You can configure tools without MCP or with it. No matter. You still have to provide that as context to an LLM.

If you're using native tool calls and not MCP, the latency of calls is a nonfactor; that was the concern raised by the root comment.