[Feature Request] Selective MCP Access Control for SubAgents

Question

**Bug Description**
I want to define MCPs that are accessible only to specific sub-agents, and not to the main agent, so the main agent’s context stays clean.
Use case: Some MCPs are meant to be used exclusively by one or more sub-agents, and should not appear in the main agent’s context.
Possible solution: Extend the main agent’s config to include an allowlist of MCPs, similar to how sub-agent configs define their MCPs.

**Environment Info**
- Platform: darwin
- Terminal: vscode
- Version: 1.0.