- 25 Excel ops (workbook/sheet/data/format/chart/pivot/validation). No Microsoft Excel required (
openpyxl). - Two ways to run: full server (stdio/SSE/HTTP) with all tool schemas in context, or the minimal path (slate docs plus
run/xl_cli.pyorrun/*.sh) where you only load one op doc and the runner's stdout.
| Full (all 25 tools) | Minimal (slate + runner) | |
|---|---|---|
| Baseline | ~2379 tokens | ~446 total, ~18 per op |
| Example (3 ops) | 293 tokens | ~54 tokens |
Minimal path uses about 5× fewer tokens for the tool surface in that scenario. Full surface = discoverability and full param docs; minimal = less context, deterministic, easier to debug. When to use which: notes/when-full-surface.md.
Full methodology, per-tool table, and how to reproduce: analysis/CONTEXT-REPORT.md.
- Python 3.10+. Paths must be absolute.
./run/create_workbook.sh /tmp/my.xlsx
./run/create_worksheet.sh /tmp/my.xlsx Sheet2
# or
python3 run/xl_cli.py create_workbook /path/to/file.xlsx
python3 run/xl_cli.py create_worksheet /path/to/file.xlsx Sheet2- Each op has a short doc in
slate/*.md. The scripts inrun/(run/<op>.shandrun/xl_cli.py) use the same logic as the server, just no server process.
- Runs over stdio, SSE, or streamable HTTP. See TOOLS.md for how to wire it up.
create_workbook |
create_worksheet |
get_workbook_metadata |
write_data_to_excel |
read_data_from_excel |
format_range |
merge_cells |
unmerge_cells |
get_merged_cells |
apply_formula |
validate_formula_syntax |
create_chart |
create_pivot_table |
create_table |
copy_worksheet |
delete_worksheet |
rename_worksheet |
copy_range |
delete_range |
validate_excel_range |
get_data_validation_info |
insert_rows |
insert_columns |
delete_sheet_rows |
delete_sheet_columns |
| Path | Purpose |
|---|---|
slate/ |
~18 tokens per op, one .md per tool |
run/ |
xl_cli.py + run/<op>.sh |
analysis/ |
token counts, baselines, CONTEXT-REPORT |
notes/ |
when full vs minimal makes sense |