MCP Server Security Checklist: 25 Controls for AI Agents
Audit MCP server security with 25 source-backed controls covering authentication, permissions, prompt injection, secrets, sandboxing, logging, revocation, and allow-or-deny tests.
Model Context Protocol turns an agent into a client of external tools, which makes discovery and authorization part of the engineering problem. These pages cover the observable failure modes—connected but empty, filtered, stale, or never called—alongside authentication, least privilege, and revocation controls.
Use the debugger guide for a minimal reproduction before changing configuration. Use the security guide and setup examples before granting credentials, network access, or write permissions to an agent workflow.
Dated product changes connected to this operating decision.
OpenAI's June 2026 Codex update turns coding-agent workflows into role-specific plugins, shareable sites, and in-place annotations.
Microsoft's Visual Studio June update makes Copilot agent mode generally available with MCP support for external tools and richer development context.
JetBrains and Zed launched an ACP Agent Registry so developers can discover and install compatible AI coding agents directly inside supported IDEs.