Building a CLI agent for Claude to manage How to design a multi-agent system where Claude delegates work?
A developer community is discussing how to design a multi-agent system where an LLM, specifically Claude, can manage another worker agent. The core question revolves around identifying the best frameworks or architectural approaches for this delegation, considering options like MCP, custom CLI wrappers, or specialized API loops. The community is seeking thoughts, setups, or recommendations on this topic.
- Published
- Sep 7, 2026, 14:49
- Source type
- Dev community
- Tier
- Community
- Source status
- Healthy
Times shown in UTC
More details
Hey everyone,
I’ve been experimenting with a terminal-based workflow where I use Codex to execute tasks and Claude to manage the overall project. Right now, I just write my project prompts or requirements directly into the terminal, and they handle everything.
However, I feel like they struggle a bit with high-level architectural design and UX flow.
I want to build my own custom AI agent that works similarly specifically, an agent that Claude itself can call, manage, and delegate tasks to via the command line.
Has anyone built a setup where Claude orchestrates a secondary sub-agent in the terminal?
What frameworks or architectural approaches (like MCP, custom CLI wrappers, or specialized API loops) are best suited for letting an LLM manage another worker agent?
Any tips on improving the design capabilities of these agents when operating in a terminal environment?
Would love to hear your thoughts, setups, or recommendations!