Investigating security issues with ChatGPT and the GitHub MCP server

This article demonstrates how GitHub’s MCP server, paired with OpenAI Codex and Sysdig alerts, can streamline investigation of a Kubernetes security alert where a job modifies /etc/hosts; it shows how the agent locates the offending code, evaluates risk, recommends Kubernetes-native fixes (e.g., hostAliases, CoreDNS, Services), and automates creating an actionable GitHub issue. It highlights practical benefits for incident response and developer workflows while noting important limitations such as inconsistent agent reasoning and vulnerability to prompt manipulation, underscoring the need for human oversight.