MCP-Guard: A Multi-Stage Defense-in-Depth Framework for Securing Model Context Protocol in Agentic AI

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Xing, Wenpeng, Qi, Zhonghao, Qin, Yupeng, Li, Yilin, Chang, Caini, Yu, Jiahui, Lin, Changting, Xie, Zhenzhen, Han, Meng
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866918277248188416
author Xing, Wenpeng
Qi, Zhonghao
Qin, Yupeng
Li, Yilin
Chang, Caini
Yu, Jiahui
Lin, Changting
Xie, Zhenzhen
Han, Meng
author_facet Xing, Wenpeng
Qi, Zhonghao
Qin, Yupeng
Li, Yilin
Chang, Caini
Yu, Jiahui
Lin, Changting
Xie, Zhenzhen
Han, Meng
contents While Large Language Models (LLMs) have achieved remarkable performance, they remain vulnerable to jailbreak. The integration of Large Language Models (LLMs) with external tools via protocols such as the Model Context Protocol (MCP) introduces critical security vulnerabilities, including prompt injection, data exfiltration, and other threats. To counter these challenges, we propose MCP-GUARD, a robust, layered defense architecture designed for LLM-tool interactions. MCP-GUARD employs a three-stage detection pipeline that balances efficiency with accuracy: it progresses from lightweight static scanning for overt threats and a deep neural detector for semantic attacks, to our fine-tuned E5-based model which achieves 96.01\% accuracy in identifying adversarial prompts. Finally, an LLM arbitrator synthesizes these signals to deliver the final decision. To enable rigorous training and evaluation, we introduce MCP-ATTACKBENCH, a comprehensive benchmark comprising 70,448 samples augmented by GPT-4. This benchmark simulates diverse real-world attack vectors that circumvent conventional defenses in the MCP paradigm, thereby laying a solid foundation for future research on securing LLM-tool ecosystems.
format Preprint
id arxiv_https___arxiv_org_abs_2508_10991
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle MCP-Guard: A Multi-Stage Defense-in-Depth Framework for Securing Model Context Protocol in Agentic AI
Xing, Wenpeng
Qi, Zhonghao
Qin, Yupeng
Li, Yilin
Chang, Caini
Yu, Jiahui
Lin, Changting
Xie, Zhenzhen
Han, Meng
Cryptography and Security
Artificial Intelligence
While Large Language Models (LLMs) have achieved remarkable performance, they remain vulnerable to jailbreak. The integration of Large Language Models (LLMs) with external tools via protocols such as the Model Context Protocol (MCP) introduces critical security vulnerabilities, including prompt injection, data exfiltration, and other threats. To counter these challenges, we propose MCP-GUARD, a robust, layered defense architecture designed for LLM-tool interactions. MCP-GUARD employs a three-stage detection pipeline that balances efficiency with accuracy: it progresses from lightweight static scanning for overt threats and a deep neural detector for semantic attacks, to our fine-tuned E5-based model which achieves 96.01\% accuracy in identifying adversarial prompts. Finally, an LLM arbitrator synthesizes these signals to deliver the final decision. To enable rigorous training and evaluation, we introduce MCP-ATTACKBENCH, a comprehensive benchmark comprising 70,448 samples augmented by GPT-4. This benchmark simulates diverse real-world attack vectors that circumvent conventional defenses in the MCP paradigm, thereby laying a solid foundation for future research on securing LLM-tool ecosystems.
title MCP-Guard: A Multi-Stage Defense-in-Depth Framework for Securing Model Context Protocol in Agentic AI
topic Cryptography and Security
Artificial Intelligence
url https://arxiv.org/abs/2508.10991