I'm curious, what issue does that solve? I'm only working on agents that make tool calls via HTTP in a home baked way but I can't imagine how resolving the tools from 2 MCP servers is harder than 1.
We've found that monolithic agents just don't perform that well.