The Story
APIs didn't go away when AI arrived. They became more important. Every agent needs to discover the right tool, call it safely, and prove the contract still holds tomorrow. Three problems. No shared stack to solve them.
MCP gave us a shared shape for what an AI-callable tool looks like. A protocol on its own doesn't catalog, govern, or test anything. That part is still bespoke glue.
So I'm building the three pieces as one stack. clictl is the catalog: a package manager that turns any API or MCP server into a tool an agent can find and call without custom integration code. SBproxy is the gateway: one runtime that governs every API, every agent, and every model. mcptest is the test runner: it puts an LLM jury in front of your MCP server, so you can tell whether the next model release will quietly break it.
Catalog, gateway, tests. One stack for the enterprise AI loop.
