MARC v1: An Open-Source Multi-Agent Framework for Clinical AI Reasoning and Coordination

MARC (Multi-Agent Reasoning and Coordination) addresses the limitations of monolithic LLM prompting in clinical reasoning, where opaque outputs and manual prompt engineering hinder adoption and debugging. Its deterministic multi-agent orchestration assigns distinct roles—extraction, reasoning, answer generation, and evaluation—to specialized agents, with explicit context passing and traceable intermediate outputs that allow stage-wise attribution of failures.

To eliminate manual prompt engineering, MARC introduces a Decomposer module that generates task-specific agent prompts from a plain-language description. The framework is entirely configurable via YAML, supports both API-based and local CPU-compatible deployments, and is designed to be model-agnostic and interpretable, making it accessible to clinical domain experts without programming expertise. The full framework is available at https://github.com/Penn-RAIL/MARC-v1.