The OWASP foundation released the OWASP MCP Top 10 (2025), the first dedicated security taxonomy for Model Context Protocol servers, vital for tools handling real-world data. The content outlines the implementation of each control in mpesa-mcp, securing Kenya's M-PESA Daraja API as an MCP server.
The author developed MCPWatch, an open-source scanner to identify security vulnerabilities in public MCP servers on GitHub. Initial results uncovered critical flaws, including CVEs, prompt injection risks, and path traversal issues, highlighting the absence of a public security registry for MCPs.
The article addresses memory poisoning as a critical, often overlooked security vulnerability in AI agents with persistent memory. This attack involves injecting malicious content into an agent's memory store, causing adversarial behavior in future sessions, a threat distinct from prompt injection.
The article explores three specific security issues in multi-agent AI systems, which arise due to implicit trust boundaries between agents. It discusses vulnerabilities documented in the OWASP Agentic AI Top 10, such as an LLM misinterpreting tool-returned data as instructions.
This tutorial addresses memory poisoning in AI agents, a critical vulnerability classified as ASI06 by OWASP. It demonstrates how to protect LangChain agents against this attack using OWASP Agent Memory Guard, an open-source runtime defense layer that screens for prompt injection and credential leakage.