MCP Protocol Vulnerabilities Lead to Data Breaches
Security researchers confirmed multiple vulnerabilities in the Model Context Protocol, including prompt injection and tool-poisoning attacks that allow malicious content to hijack AI agent behavior. The NSA documented a remote code execution flaw in the MCP-Inspector toolchain and described semantic tool-poisoning as systemic. Tencent's Zhuque Lab released AI-Infra-Guard, an open-source framework for auditing MCP servers and third-party agent skill packages as a supply chain.
What happened
Security researchers confirmed in April 2025 that the Model Context Protocol (MCP) carries multiple outstanding vulnerabilities, including prompt injection and tool-poisoning attacks that allow malicious content to hijack agent behavior through ordinary tool interactions.
A National Security Agency advisory published in May 2026 documented CVE-2025-49596, a remote code execution flaw in the MCP-Inspector toolchain, and described semantic tool-poisoning attacks as systemic rather than isolated.
Tencent's Zhuque Lab released AI-Infra-Guard on June 30, 2026, an open-source framework that applies deterministic rules to infrastructure and LLM auditing to MCP servers and third-party agent skill packages as a supply chain.
Reported impact
MCP vulnerabilities sit at the center of enterprise agentic deployments, as MCP is the de facto interface standard adopted by Anthropic, OpenAI, Google DeepMind, and dozens of enterprise tooling vendors for connecting AI agents to external tools.
Index analysis
Existing red teaming tools such as Microsoft's PyRIT, NVIDIA's Garak, and Promptfoo target only the model layer—the direct interface between users and the underlying language model—and are not adequate for testing agents that browse websites, read files, call APIs, write code, and execute it.
AI-Infra-Guard addresses vulnerabilities in the tool integration layer by testing MCP servers for adversarial content and auditing third-party agent skill packages.
Governance controls
Tencent released AI-Infra-Guard as an open-source framework to enable organizations to audit MCP infrastructure and detect vulnerabilities in their AI agent deployments.
Classification
Sources and evidence
This record was researched and written by the Index, based on the following original reporting.