A simple 19-character AI prompt triggered a 118,693-byte request payload

A developer discovered that typing a 19-character question into Claude Code generated a network request nearly 6,200 times larger than the message itself. The payload was dominated by 38 tool JSON schemas totalling over 78,000 bytes, a lengthy system prompt, and deferred tool listings — none of which were actually used to answer the question. Notably, over 44,000 bytes of the tool schemas consisted of English prose documentation rather than machine-readable data, with one tool alone carrying a 19,000-byte description. Sixteen of the 38 bundled tools were for controlling a browser preview the user had never opened, and a dedicated tool existed solely to load even more tools. The audit highlights how AI coding assistants silently transmit large, largely irrelevant context on every single request, regardless of query complexity.