Before the Agent Calls: Source-level Findings from 100 MCP Servers

Before the Agent Calls: Source-Level Findings from 100 MCP Servers MCP security isn't just about what happens at runtime. In many cases, the vulnerabilities are already sitting in the server's source code before an agent ever makes its first tool call. In this session, Akash Sathish, Solution Consultant at Sahaj Software, shares findings from an AST-level audit of 100 of the most-installed TypeScript MCP servers, uncovering 445 real security findings across 70 servers. The talk examines vulnerabilities including command injection, path traversal, SSRF, and tool poisoning, showing how these issues can make it into production code and why traditional review, and even LLM-based scanners, can miss them. Akash also demos MCPeek, an open-source static analysis tool he built using ts-morph. The scanner performs multi-pass taint tracking from tool inputs to dangerous sinks, maps findings to the OWASP MCP Top 10, and can emit SARIF results directly into GitHub Code Scanning — all fully offline, with no model or cloud dependency. The session also takes an honest look at the limits of static analysis and where dynamic and runtime security approaches still have an advantage. 🎤 Speaker: Akash Sathish - Solution Consultant, Sahaj Software 📌 Topic: Before the Agent Calls Source-Level Findings from 100 MCP Servers 🔗 MCPeek: https://github.com/iamakash-06/MCPeek 📦 NPM Package: https://www.npmjs.com/package/mcpeek Want to be part of the community building the future of agentic AI? Join the Agentic AI Foundation (AAIF) and connect with the people, projects, and organizations shaping open standards for agentic AI. https://aaif.io/news/linux-foundation-announces-formation-of-aaif