Scored across 237 tools
Descriptions are unusually diligent about cross-referencing alternatives (e.g. "use list_missing_submissions instead", "get_rubric_assessment returns the same data plus..."), which rescues many pairs. But at 237 tools there is real overlap: daily_briefing vs get_week_ahead vs semester_snapshot vs course_health_check, get_activity_stream vs get_all_unread vs diff_since_last_check, get_outcome_rollup vs get_my_outcome_results, and a long tail of mark_* and list_upcoming/list_todo/list_planner variants that an agent could easily misselect.
The dominant pattern is snake_case verb_noun (list_courses, get_assignment, submit_assignment_file, mark_entry_read), applied consistently with no camelCase mixing. There is a noticeable minority of noun-first composite names (daily_briefing, assignment_brief, semester_snapshot, course_health_check, grade_impact_ranking, whoami, grep_all_courses), which is a minor deviation rather than a breakdown.
237 tools is far beyond what an agent can select from reliably; the server itself ships search_tools to help find things "among the 200+ available", which is an admission the surface is unmanageable. Even granting Canvas's broad domain, many tools are narrow conveniences (get_todo_count, get_storage_quota, mark_conversation_unread, dismiss_global_announcement) that inflate the surface rather than earn a slot.
Coverage across courses, assignments, submissions, grading, quizzes, discussions, files, modules, planner, calendar, groups, outcomes, peer review and export is close to exhaustive for a student-facing Canvas client, and raw_canvas_request provides an escape hatch for anything missed. Gaps are deliberate or platform-limited (quiz taking is explicitly never supported, ePortfolio editing is not possible via API, no instructor grading writes).