Go to TMLR homepageCross-Server Interoperability in Multi-MCP Automated AI Agent Networks
Abstract: This paper introduced a combined framework for cross-server interoperability in multi-MCP
automated AI agent networks. The design combined communication abstraction, orchestra-
tion optimization, and security validation. The framework was tested on BoT-IoT, ToN-IoT,
and PettingZoo datasets, which represented adversarial traffic detection, telemetry-heavy
IoT environments, and dynamic multi-agent orchestration. Results showed improvements
in coverage, efficiency, and robustness, with accuracy, precision, recall, and F1-score above
0.95 across multiple trials. Ablation analysis confirmed the role of each component, scalabil-
ity tests showed stable performance as servers increased, and stress evaluations demonstrated
graceful degradation under heavier attack loads. Error analysis and statistical validation
supported the reliability of the outcomes, while resource usage comparisons indicated re-
duced runtime and memory consumption against baselines. Cross-domain generalization
confirmed adaptability across unseen datasets. These findings demonstrated that inter-
operability in heterogeneous MCP networks can be achieved without sacrificing efficiency,
scalability, or reliability, providing a foundation for secure and practical multi-domain agent
collaboration
Submission Length: Long submission (more than 12 pages of main content)
Changes Since Last Submission: > Summary Of Contributions: This paper proposes a combined framework for multi-MCP automated AI agent network to address the cross-server interoperability. This paper introduces and explains the problem well, and covers the related work extensively. However, the main structure of the algorithm was rather difficult to follow, and the experiment results was not well explained, making it difficult to assess the contribution of the work.
Response to Reviewer: We thank the reviewer for the constructive feedback. In response to the comment regarding the clarity of the algorithm structure and the explanation of experimental results, we have revised the manuscript accordingly.
Clarified Contributions (Page 3): A concise “Summary of Contributions” paragraph has been added at the end of the Introduction. This explicitly outlines the major contributions of the framework and improves the overall readability of the paper.
Improved Algorithm Explanation (Page 12): At the start of the Results and Analysis section, we added a short explanatory paragraph that maps each experimental metric to its corresponding component of the proposed framework. This helps readers understand how the evaluation supports the main contributions.
Strengthened Results Contribution Link (Pages 13–15): A clarifying sentence has been added at the end of each dataset subsection (BoT-IoT, ToN-IoT, and PettingZoo). These additions explicitly connect the outcomes of each experiment to the contribution of the proposed framework.
> Are the claims made in the submission supported by accurate, convincing and clear evidence?: No
> Explain your answer above: While the paper explained the problem well, the proposed methodology was difficult to follow. It was difficult to understand what the algorithm is doing from the Figure 1 alone. For example, PetingZoo, BoT-IoT and ToN-IoT are evaluation benchmarks, not what this algorithm is supposed to solve in real life.There are also a lot of names, nomenclatures, and variable names that could not be understood by looking at the Figure 1 alone without reading the following sections.
Response: We replaced Figure 1 with a clearer top-to-bottom conceptual architecture, removed datasets from the diagram, and simplified the module structure to avoid unnecessary definitions. We also added a short clarifying paragraph after the figure to explain that it presents a high-level overview rather than the full algorithm, and to show how the detailed steps are provided in Sections 3.1–3.6. The surrounding methodology text was streamlined to align with the updated figure. These revisions resolve the concerns about clarity, dataset placement, and terminology.
> There were some other details in the main text that made it difficult to understand the algorithm. For example, in the equations 3 and 9. In the equations, the algorithm uses function f and H to secure Messages. Are messages in this case generated by simple concatenation or by using an embedding network? If former, using f and H might be misleading as they denote actual function, not a group of element.
Response: To remove ambiguity around the use of the functions f(⋅)and H(⋅)in Equations (3) and (9), we added clarifying sentences on page 9 stating that f(⋅)represents simple concatenation and H(⋅)denotes a standard cryptographic hash (e.g., SHA-256), rather than learned functions. We also added a short clarification on page 10 indicating that all functions in the security model follow conventional cryptographic definitions. These additions resolve the confusion regarding message generation and hashing behavior.
> Some of the tables exceeds page limit, and some of the texts extend beyond their cells. While most cases are OK, there are some cases where the formatting error is too severe that it is impossible to read the table, such as Tables 7 and 8.
Response: We have corrected both tables 7 and 8 by resizing them to fit within the page width, adjusting column types to prevent text overflow, and ensuring proper alignment. Table 7 on page 16 and Table 8 on page 20 now appear fully readable and do not exceed page limits. We confirm that the manuscript contains eight figures in the main text, consistent with TMLR expectations, and two supplementary figures placed in the appendix to avoid interrupting the narrative flow while still providing full experimental transparency.
Assigned Action Editor: ~Jeff_Phillips1
Submission Number: 6207
Loading