arXiv cs.AIThursday · May 28, 2026FREE

TCP-MCP: Landscape-Guided Co-Evolution of Prompts and Communication Topologies for Multi-Agent Systems

TCP-MCP (Topology-Coupled Prompting for Multi-Agent Collaborative Problem-Solving) is a co-evolution framework that treats agent prompts and communication topologies as a unified genome. It uses an initialization-time landscape probe to calibrate early search behavior and Pareto-front diagnostics to adapt exploration under three objectives: task performance, token cost, and structural complexity. Using DeepSeek-V3.2 as the backbone across all methods, TCP-MCP achieves 82.66% accuracy on MMLU-Pro, 89.96% on MMLU, and 96.61% on GSM8K. It consistently outperforms automated graph-generation baselines and achieves competitive accuracy relative to debate-style systems, while using up to 5.69× fewer tokens at the reported operating points. The paper is available on arXiv (2605.27850v1).