Tool definitions are not function signatures. They are menu items an LLM reads to decide what to call and how to fill the arguments. This post covers what separates good tool definitions from bad ones, with before/after examples and a practical checklist.
Tool quality directly affects agent reliability. Arcade Evals lets you test MCP tool definitions against LLM behavior in simulated conversations, scoring whether the model selects the right tool and populates arguments correctly, before you deploy to production.
Picture this: You walk into a newly opened restaurant for the first time, excited by the positive reviews, and confidently stride to a window-side table. As soon as you're settled, the waiter approaches, but before they can speak, you say, "The usual, please." The waiter stares at you like you've lost your mind. They've never seen you before. They have no idea what "the usual" means. That's your AI agent every time you start a new conversation. It has absolutely no idea who you are, what you w
