Will an agent actually understand your tool?

An LLM decides whether and how to call a tool almost entirely from its name, description and JSON-Schema params — a tool definition is a data contract for agents. Tool Forge lints that contract the way a model reads it: ambiguous descriptions, untyped or loose params, missing required, boolean traps — then previews the tool card and a sample agent call. All client-side.

Why this matters. Most agent failures aren't model failures — they're contract failures: a vague description, an untyped param, a boolean nobody can read. The discipline is identical to the data contracts behind a national dataset: make the schema say exactly what's allowed, and validate it before anything depends on it. The same lint could run in CI on every MCP server. Built by John Mikel Regida — companion to zeno-mobile-runner (an MCP test runner).