The reason agents are exploding right now is that we figured out how to give them access to tools. MCP is a huge part of this. But nobody is talking about how to secure these servers. The lack of security is making people throw their computers out the window. If you're building an MCP server, you have two choices: 1. Use static API keys → This is a disaster. 2. Become an expert in OAuth 2.1, PKCE, and the MCP spec. Neither option is great. Agents are fundamentally different from humans: • They should be able to act autonomously • They should be able to work across multiple systems • They should have dynamic access at the function and tool levels • Their credentials should be short-lived • Their credentials should be scoped as tightly as possible Traditional IAM systems don't work for this. Fortunately, we now have a solution!