Related Servers

Several tool clusters have genuinely unclear boundaries: get_data_validation_info vs inspect_data_validation_rules, append_table_rows vs append_excel_table_rows, and a four-way read overlap (read_data_from_excel, read_excel_as_table, quick_read, query_table) that only the verbose descriptions partially untangle. Multiple write paths (write_data_to_excel, apply_workbook_changeset, apply_table_patch, upsert_excel_table_rows, update_rows_by_key) also target overlapping mutations, so tool selection is error-prone despite detailed prose.

Everything is snake_case and mostly verb_noun, but the read verbs are scattered across read_/get_/inspect_/list_/describe_/profile_/audit_ with no rule distinguishing them (get_merged_cells vs inspect_named_range vs list_named_ranges). A few names invert the pattern (cross_workbook_lookup, search_in_sheet, freeze_panes, bulk_filter_workbooks), keeping it readable but not predictable.

83 tools is far beyond what a coherent Excel surface needs, and much of the bulk is near-duplicate read/write/validation variants rather than distinct capabilities. The set reads as accumulated feature creep, making it hard for an agent to discover the canonical tool for a task.

Coverage of the Excel domain is unusually thorough: reading, writing, formatting, tables, charts, pivots, named ranges, protection, print setup, validation/conditional-format inspection, formula analysis, recipes, patches, snapshots, and diffs. Missing pieces exist (creating conditional formatting or data validation rules, sorting, chart/pivot deletion or update, worksheet reordering, workbook-level protection), but agents can work around most of them.