[Feature Request] MCP Tool Visibility Scoping for Main Context vs Subagents

Bug Description
Feature Request: MCP Tool Visibility Scoping (Main Context vs Subagents)

Problem:
When configuring MCP servers (especially via plugins), all exposed tools appear in the main conversation context, contributing to context overhead. For projects with multiple MCP servers exposing many tools, this significantly impacts the available context window even when those tools are only intended for use by specialized subagents.

Desired Behavior:
Add a configuration option to control MCP tool visibility:

• "visibility": "subagent-only" - Tools available only when spawned as/within a subagent
• "visibility": "main-only" - Tools available only in main context
• "visibility": "all" (default) - Current behavior

Example Configuration:
{
"mcpServers": {
"code-explorer": {
"command": "node",
"args": ["server.js"],
"visibility": "subagent-only"
}
}
}

Use Case:
I have plugins with MCP servers (semantic search, LSP tools) that are designed to be used exclusively by specialized subagents. The tools take up context in the main thread but are never directly invoked there. Being able to scope their visibility would:

1. Reduce context overhead in the main conversation
2. Enforce architectural patterns (tools only via subagents)
3. Keep the main tool list cleaner and more focused

Environment Info

• Platform: darwin
• Terminal: cursor
• Version: 2.0.76
• Feedback ID: 653583ea-51ac-46d1-a9a5-f933946db73b

Note: Error logs were truncated.