Scored across 72 tools
Multiple tool families overlap heavily: get_dependencies/get_dependents/get_change_impact/get_full_context/get_edit_context all analyze symbol relationships; search_codebase/search_in_symbols/ts_search all search code; memory_search/reasoning_search/capture_search all search stores. Descriptions help somewhat, but boundaries remain unclear and agents must memorize subtle distinctions to avoid misselection.
Nearly all tools use snake_case with predictable verb_noun patterns and consistent family prefixes (get_, memory_, capture_, reasoning_, ts_, corpus_). Minor deviations like 'checkpoint' (noun-only) and 'reindex' are readable but slightly inconsistent.
72 tools is far beyond the typical 3-15 range and meets the rubric's extreme mismatch threshold. Even for a broad code-intelligence platform, this creates an overwhelming selection burden and increases the chance of wrong tool choice.
The surface covers code navigation, editing, refactoring, dependency analysis, git status, diagnostics, memory, reasoning, captures, and project management comprehensively. Only minor gaps exist, such as dedicated file creation/deletion and git commit/push operations, but major workflows are supported.