Executor SDK ships lazy-loaded tool discovery

The hot take is right: too many tools is not the problem. Bad discovery, bad auth, and bad policy boundaries are the problem, and Executor is trying to solve those three layers together.

• –It treats MCP, OpenAPI, GraphQL, and custom sources as one indexed catalog, so agents can discover capabilities by intent rather than memorized paths
• –Shared auth and policy handling matter more than raw tool count; that is what makes broad access practical without turning every call into a permission headache
• –The SDK is really an execution layer, not just a helper library, which makes it more useful for teams standardizing multiple agents and tool backends
• –The upside scales with catalog quality: if schemas are clean and search is good, lots of tools becomes a feature; if not, it becomes a maze
• –The local-first design is a strong fit for developer workflows that want power without shipping secrets into every agent integration