Turn on the Tool Search Tool with defer_loading to stop paying for tool definitions you don't use

Anthropic Engineering·high signal

Mark rarely-used tools with "defer_loading": true and add a {"type": "tool_search_tool_regex_20251119"} tool so Claude discovers definitions on demand instead of loading all of them upfront; keep only 3–5 high-use tools always loaded. Anthropic's numbers are large: 58 tools drop from ~55K to ~8.7K tokens (85% cut), and Opus 4.5 tool-selection accuracy rises from 79.5% to 88.1%. For a builder running MCP servers with dozens of tools, this is the single highest-leverage change to reclaim context and improve routing accuracy.