Exa Search: tool descriptions scored

Scores, issues and names only

This page shows how Exa Search's tool descriptions scored, the issues found and the names of its tools and parameters. The descriptions are the server's own: read them where the copy came from.

This copy comes from Smithery's registry, which keeps each tool's name, description and input schema, but no output schema, annotations or instructions. Points for return values and for marking destructive tools can come out lower than the server's own tools/list would give.

A server with a public URL can be scored as it is today: score from a URL

What static scoring cannot tell

This score reads only the text of each description and schema: it says what is written, not what a model will do. Short tools whose names already say what they do, such as browser_close, score low here even when models use them well, and in our measurements current models chose the right tool in one step even from one-line descriptions, as long as the names and schemas differed. What the text cannot show, such as what happens over several steps and whether an added sentence helps or misleads, is what an evaluation measures, on the Pro and Team plans.