大語言模型（LLMs）的快速發展推動了多智能體系統（Multi-Agent Systems, MAS）的典範轉移。然而，在處理複雜工作流時，整合高度專業化且異構的 MAS 往往依賴僵硬的「膠水代碼（Glue Code）」。這種強耦合的傳統整合方式導致了嚴重的通訊瓶頸，並大幅限制了系統的可擴展性與開發敏捷性。
為解決此系統性瓶頸，本研究提出並實作了一個「協定驅動之異構多智能體平台（Protocol-Driven Heterogeneous MAS Platform, PDP）」。本平台採用無框架依賴的 Agent-to-Agent (A2A) 協定與創新的「MAS Card」元數據架構，搭配 Docker 容器化技術，成功解決了異構系統在通訊、邏輯與層級，以及軟體基礎設施上的三大異質性問題。本研究以量化投資領域作為具體驗證場景，成功無縫整合了兩個獨立開發的多智能體系統：Alpha 挖掘智能體系統（AMiAS）與 Alpha 建模智能體系統（AMoAS）。
為了嚴謹驗證平台效能，本研究設計了涵蓋六大評估維度的 27 組測試資料集。實
驗結果顯示，中央門戶智能體（Portal Agent）在任務意圖識別與動態派發上達到 100%的準確率，並能有效執行複合任務的平行拆分。在邊界防禦與系統韌性上，Portal Agent 成功扮演了嚴格的「守門員（Gatekeeper）」角色，不僅能主動攔截領域外請求，更能依賴完備的元數據防止下游系統產生「靜默失敗（Silent Failures）」。此外，透過 BAAI/bge-m3 模型進行向量相似度分析，證實了意圖在轉譯與傳遞過程中具有極高的語義保真度。最後，G-EVAL 使用者滿意度評估揭示了本架構的延伸貢獻：平台具備優異的「輸出異質性（Output Heterogeneity）」適應力，能動態針對生硬的量化機器產出進行文本去噪與易讀性補償，同時對高品質報告維持無損穿透。
本研究證實，協定驅動架構能有效取代僵硬的腳本整合，為未來去中心化、大規模異構 MAS 的協作與擴展，提供了一個具備高度彈性、穩定性與一致性體驗的基礎框架。

The rapid evolution of Large Language Models (LLMs) has catalyzed a paradigm shift toward Multi-Agent Systems (MAS). However, orchestrating highly specialized, heterogeneous MAS to tackle complex workflows often relies on rigid, hardcoded ”glue code.” This tightly coupled approach creates severe communication bottlenecks and limits the scalability and agility of system development.
To address these systemic bottlenecks, this research proposes and implements a Protocol-Driven Heterogeneous MAS Platform (PDP) for heterogeneous MAS. By leveraging a framework-agnostic Agent-to-Agent (A2A) protocol, a novel ”MAS Card” metadata schema, and Docker containerization, the platform effectively resolves communication, logical, hierarchical, and software infrastructural heterogeneities. Operating within the domain of quantitative investment as a rigorous testbed, the platform seamlessly integrates two independently engineered systems: the Alpha Mining Agent System (AMiAS) and the Alpha Modeling Agent System (AMoAS).
The platform’s architecture is comprehensively evaluated across six core dimensions using a specialized 27-query dataset. Empirical results demonstrate a flawless 100% success rate in task intent recognition and dynamic dispatching by the central Portal Agent. Furthermore, the platform successfully executes parallel task splitting for composite queries. In terms of boundary defense and system resilience, the Portal Agent acts as an effective gatekeeper —intercepting out-of-domain requests and relying on absolute metadata completeness to prevent downstream silent failures and logical hallucinations. Vector similarity analysis using the BAAI/bge-m3 model confirms exceptionally high task intent fidelity, proving that instructions are routed with zero semantic distortion. Crucially, the reference-free G-EVAL satisfaction assessment reveals an emergent architectural contribution: the platform＇s exceptional adaptability to ”output heterogeneity.” The Portal Agent dynamically provides value-added text refinement for raw quantitative logs while switching to high-fidelity transparent transmission for natively polished reports.
This research validates that a protocol-driven architecture successfully eliminates the reliance on rigid scripted integrations, providing a highly scalable, resilient, and standardized foundation for orchestrating decentralized and heterogeneous MAS ecosystems.

[1]      A2A Project, “A2A,” GitHub repository, 2025. [Online]. Available: https://github. com/a2aproject/A2A.
[2]      S. An, Q. Li, J. Lu, D. Yin, and X. Sun, “FinVerse: An Autonomous Agent System for Versatile Financial Analysis,” arXiv preprint arXiv:2406.06379, 2024.
[3]      J. Chen, S. Xiao, P. Zhang, K. Luo, D. Lian, and Z. Liu, “M3-Embedding: Multi-Linguality, Multi-Functionality, Multi-Granularity Text Embeddings Through Self-Knowledge Distillation,” in Findings of the Association for Computational Linguistics: ACL 2024, L.-W. Ku, A. Martins, and V. Srikumar, Eds. Bangkok, Thailand: Association for Computational Linguistics, Aug. 2024, pp. 2318–2335.
[4]      C. Cui, W. Wang, M. Zhang, G. Chen, Z. Luo, and B. C. Ooi, “AlphaEvolve: A Learning Framework to Discover Novel Alphas in Quantitative Investment,” in Proceedings of the 2021 International Conference on Management of Data, 2021, pp. 2208–2216.
[5]      A. Ehtesham, A. Singh, G. K. Gupta, and S. Kumar, “A Survey of Agent Interoperabil-ity Protocols: Model Context Protocol (MCP), Agent Communication Protocol (ACP), Agent-to-Agent Protocol (A2A), and Agent Network Protocol (ANP),” arXiv preprint arXiv:2505.02279, 2025.
[6]      S. Es, J. James, L. Espinosa Anke, and S. Schockaert, “RAGAs: Automated Evaluation of Retrieval Augmented Generation,” in Proceedings of the 18th Conference of the European Chapter of the Association for Computational Linguistics: System Demonstrations, St. Julians, Malta, Mar. 2024, pp. 150–158.
[7]      M. Gao, Y. Li, B. Liu, Y. Yu, P. Wang, C.-Y. Lin, and F. Lai, “Single-Agent or Multi-Agent Systems? Why Not Both?,” arXiv preprint arXiv:2505.18286, 2025.
[8]      A. Gulli, L. Nigam, J. Wiesinger, V. Vuskovic, I. Sigler, I. Nardini, N. Stroppa,
S. Kartakis, N. Saribekyan, A. Nawalgaria, and A. Bount, “Agent Compan-ion,” Kaggle Whitepaper, 2024. [Online]. Available: https://www.kaggle.com/ whitepaper-agent-companion.
[9]      T. Guo, X. Chen, Y. Wang, R. Chang, S. Pei, N. V. Chawla, O. Wiest, and X. Zhang, “Large Language Model Based Multi-Agents: A Survey of Progress and Challenges,” arXiv preprint arXiv:2402.01680, 2024.
[10]      J. He, C. Treude, and D. Lo, “LLM-based multi-agent systems for software engineering: Literature review, vision, and the road ahead,” ACM Trans. Softw. Eng. Methodol., vol. 34, no. 5, Art. no. 124, May 2025.
[11]      X. Hou, Y. Zhao, S. Wang, and H. Wang, “Model Context Protocol (MCP): Landscape, Se-curity Threats, and Future Research Directions,” arXiv preprint arXiv:2503.23278, 2025.
[12]      LangChain,           “Choosing   the   Right   Multi-Agent   Architecture,” 2024.   [Online].   Available:                              https://www.langchain.com/blog/ choosing-the-right-multi-agent-architecture.
[13]      LangChain AI, “LangGraph,” GitHub repository, 2024. [Online]. Available: https:// github.com/langchain-ai/langgraph.
[14]      A. Li, Y. Xie, S. Li, F. Tsung, B. Ding, and Y. Li, “Agent-Oriented Planning in Multi-Agent Systems,” arXiv preprint arXiv:2410.02189, 2025.
[15]      M. Li, Y. Zhao, B. Yu, F. Song, H. Li, H. Yu, Z. Li, F. Huang, and Y. Li, “API-
Bank: A Comprehensive Benchmark for Tool-Augmented LLMs,” arXiv preprint arXiv:2304.08244, 2023.
[16]      Q. Li and Y. Xie, “From Glue-Code to Protocols: A Critical Analysis of A2A and MCP Integration for Scalable Agent Systems,” arXiv preprint arXiv:2505.03864, 2025.
[17]      C. C. Liao, D. Liao, and S. S. Gadiraju, “AgentMaster: A Multi-Agent Conversational Framework Using A2A and MCP Protocols for Multimodal Information Retrieval and Analysis,” arXiv preprint arXiv:2507.21105, 2025.
[18]      Y. Liao, S. Jiang, Y. Wang, and Y. Wang, “Reflectool: Towards Reflection-Aware Tool-Augmented Clinical Agents,” in Proceedings of the 63rd Annual Meeting of the Associa-tion for Computational Linguistics (Volume 1: Long Papers), 2025, pp. 13507–13531.
[19]      The Linux Foundation, “A2A Protocol,” 2026. [Online]. Available: https:// a2a-protocol.org/latest/.
[20]      Y. Liu, D. Iter, Y. Xu, S. Wang, R. Xu, and C. Zhu, “G-Eval: NLG Evaluation Using GPT-4 with Better Human Alignment,” arXiv preprint arXiv:2303.16634, 2023.
[21]      Y. Liu, D. Iter, Y. Xu, W. Shuohang, R. Xu, and C. Zhu, “G-Eval: NLG Evaluation using Gpt-4 with Better Human Alignment,” in Proceedings of the 2023 Conference on Empir-ical Methods in Natural Language Processing, H. Bouamor, J. Pino, and K. Bali, Eds. Singapore: Association for Computational Linguistics, Dec. 2023, pp. 2511–2522.
[22]      N. Muennighoff, N. Tazi, L. Magne, and N. Reimers, “MTEB: Massive Text Embedding Benchmark,” in Proceedings of the 17th Conference of the European Chapter of the As-sociation for Computational Linguistics, A. Vlachos and I. Augenstein, Eds. Dubrovnik, Croatia: Association for Computational Linguistics, May 2023, pp. 2014–2037.
[23]      J. Ruan, Y. Chen, B. Zhang, Z. Xu, T. Bao, G. Du, S. Shi, H. Mao, Z. Li, and X. Zeng, “TPTU: Large Language Model-Based AI Agents for Task Planning and Tool Usage,” arXiv preprint arXiv:2308.03427, 2025.
[24]      A. Taubenfeld, T. Sheffer, E. Ofek, A. Feder, A. Goldstein, Z. Gekhman, and G. Yona, “Confidence Improves Self-Consistency in LLMs,” in Findings of the Association for Computational Linguistics: ACL 2025, 2025, pp. 20090–20111.
[25]      S. Wang, H. Yuan, L. Zhou, L. Ni, H.-Y. Shum, and J. Guo, “Alpha-GPT: Human-AI Inter-active Alpha Mining for Quantitative Investment,” in Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing: System Demonstrations, 2025,
pp. 196–206.
 [26]      J. Wei, X. Wang, D. Schuurmans, M. Bosma, F. Xia, E. Chi, Q. V. Le, and D. Zhou, “Chain-of-Thought Prompting Elicits Reasoning in Large Language Models,” Advances in Neural Information Processing Systems, vol. 35, pp. 24824–24837, 2022.
[27]      J. Wiesinger, P. Marlow, and V. Vuskovic, “Agents,” Kaggle Whitepaper, 2024. [Online]. Available: https://www.kaggle.com/whitepaper-agents.
[28]      Q. Wu, G. Bansal, J. Zhang, Y. Wu, B. Li, E. Zhu, L. Jiang, X. Zhang, S. Zhang, and J. Liu, “AutoGen: Enabling Next-Gen LLM Applications via Multi-Agent Conversation,” arXiv preprint arXiv:2308.08155, 2023.
[29]      F. Xing, “Designing Heterogeneous LLM Agents for Financial Sentiment Analysis,” ACM Transactions on Management Information Systems, vol. 16, no. 1, pp. 1–24, Feb. 2025.
[30]      H. Yang, B. Zhang, N. Wang, C. Guo, X. Zhang, L. Lin, J. Wang, T. Zhou, M. Guan, and R. Zhang, “FinRobot: An Open-Source AI Agent Platform for Financial Applications Using Large Language Models,” arXiv preprint arXiv:2405.14767, 2024.
[31]      S. Yao, D. Yu, J. Zhao, I. Shafran, T. Griffiths, Y. Cao, and K. Narasimhan, “Tree of Thoughts: Deliberate Problem Solving with Large Language Models,” Advances in Neu-ral Information Processing Systems, vol. 36, pp. 11809–11822, 2023.
[32]      Y. Yu, H. Li, Z. Chen, Y. Jiang, Y. Li, J. W. Suchow, D. Zhang, and K. Khashanah, “Fin-Mem: A Performance-Enhanced LLM Trading Agent with Layered Memory and Charac-ter Design,” IEEE Transactions on Big Data, 2025.
[33]      H. Yuan, S. Wang, and J. Guo, “Alpha-GPT 2.0: Human-in-the-Loop AI for Quantitative Investment,” arXiv preprint arXiv:2402.09746, 2024.
[34]      T. Zhang, V. Kishore, F. Wu, K. Q. Weinberger, and Y. Artzi, “BERTScore: Evaluating Text Generation with BERT,” arXiv preprint arXiv:1904.09675, 2020.
[35]      T. Zhang, Y. Li, Y. Jin, and J. Li, “AutoAlpha: An Efficient Hierarchical Evolution-ary Algorithm for Mining Alpha Factors in Quantitative Investment,” arXiv preprint arXiv:2002.08245, 2020.