Eval the command surface against the catalog #1910

Part of #1357. Plan: cli-first-mcp task 8.1 (design, tasks). Depends on #1890, #1891, #1899.

Run the full mcp-server-evals suite on the catalog, search mode, and the all-tools mode, with at least two clients or models. Record task success, tokens, and turns in docs/testing/. This also decides each client's default mode and whether Code Mode (#1670) is needed.

Done when

• Results are checked in, with a recommendation for the default.

via @BYK.

--

View Session in Sentry