PRODUCTION SCENARIO
An assistant aggregates seven MCP servers with roughly 310 tools, whose definitions consume about 62,000 tokens per request. Tool selection accuracy has dropped as the catalog grew.
Which change addresses both problems?
Answering here is anonymous. Nothing is saved unless you sign in.
Show answer and explanation
Answer: Add a tool search tool and defer all but a few frequent tools
Selection accuracy degrades past 30 to 50 available tools, and deferring definitions so Claude loads them on demand cuts definition tokens by over 85 percent while keeping accuracy high, which is why the 3 to 5 most-used tools plus the search tool stay loaded. Shortening descriptions trims tokens but makes selection worse, since detailed descriptions are what Claude selects on. Splitting into seven conversations moves the selection problem onto the user. And deferring every tool, search tool included, returns a 400: at least one must stay non-deferred.