Unverified50% confidenceEventExact time
Invariant Labs coined the term for MCP tool poisoning attacks in April 2025.
1
Sources
50%
Confidence
Long-term
Relevance
10/5/2026
First Seen
Sources
Related Entities
Related Claims
Unverified当前MCP生态缺乏类似App Store审核机制的准入管控，安全研究人员已发现多个MCP投毒攻击案例，包括Typosquatting钓鱼攻击和伪装成Postmark MCP的恶意服务器65% similarUnverified截至 2025 年，Cursor、Claude Desktop 等主流 AI 工具已接入 MCP 生态，LangGraph 通过 langchain-mcp-adapters 包提供原生支持65% similarUnverified在MCP层可以注入恶意工具返回值、拦截审计Agent的工具调用、模拟工具失败等边界情况，使MCP层成为对抗测试的注入点与观测窗61% similarVerifiedMCP tools are essentially arbitrary code execution, indirect prompt injection is a real-world threat, and attack vectors can be hidden in resource content rather than in direct user input.60% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/977555API
curl https://kongchang.com/api/v1/knowledge/claims/977555MCP
get_claim(id=977555)