Hierarchical Claude orchestration: Is our MCP bridge pattern optimal?

[michael-abdo]

Pre-submission Checklist

• This post follows the MCP community guidelines

What would you like to share?

We're building a hierarchical orchestration system with multiple Claude instances via tmux (executive→manager→specialist). Since MCP's 1:1 stdio architecture prevents spawned instances from accessing the primary's MCP tools, we implemented a bridge pattern.

• Single shared MCP server process (~50-70MB)
• Lightweight Node.js bridge that executives call via Bash
• Centralized state management with no race conditions
• ~85% memory savings vs spawning separate MCP servers per instance

Is there a more efficient pattern for multi-instance MCP access than our bridge approach, or does this represent current best practice for distributed Claude architectures?

Relevant Links

No response