Show HN: Sub-Agent MCP: LLM delegation and sub-agent orchestration via MCP (github.com)

A new production-ready Python server called Sub-Agent MCP has been announced for managing LLM delegation and orchestrating sub-agents. This server allows a parent LLM, such as Cursor, to connect and specify tasks for various specialized sub-agents through a YAML-defined configuration. Each sub-agent is linked to an individual OpenAI-compatible LLM and can access optional downstream tool servers, ensuring clear roles and lightweight parent operations. The significance of Sub-Agent MCP lies in its ability to streamline the delegation of tasks among LLMs while minimizing contextual overload for the parent agent. By implementing a structured approach using YAML for agent configuration and strict Pydantic validation, developers can easily customize the modeling details for various roles. Key features include support for streaming HTTP communication, robust logging, and production-oriented practices like Docker compatibility and CI pipeline integration. This setup enhances security by preventing the exposure of API keys and allows for flexible interactions with multiple LLMs without bloating the parent’s context, making it a valuable addition to the AI/ML development ecosystem.