Loading ToolBench
An MCP server for controlling Excel files through natural language prompts using OpenAI's GPT-4o model. Provides tools for reading, writing, and manipulating Excel spreadsheets.
Critical gaps across all dimensions. All 7 tools have empty parameter descriptions (violates pattern:tool-description). Schemas present but descriptions are minimal (10-60 chars, below 194-char baseline). No output schemas documented. Error handling is minimal, most functions return raw exceptions or generic messages. No input validation, no recovery guidance. Tool names are acceptable (verb_noun pattern) but lack context. This server would not pass code review.
Create a new sheet, create file if it doesn't exist.
Delete a sheet.
Insert values from start_value to end_value in a column, starting at start_cell.
List all sheet names in an Excel file.
Read a value from a specific cell.
Save the Excel file.
Write a value to a specific cell. Create sheet if missing. Handle spaces, cases safely.
All parameter descriptions are empty strings. Pattern:tool-description requires non-empty descriptions for every input parameter so LLMs can infer parameter meaning. Example: 'filepath' in all tools has description: ''.
No output schemas documented. Pattern:tool requires documented return types so agents know what fields to expect and can plan downstream calls. Functions return dicts but structure is not declared in tool metadata.
Error handling lacks recovery guidance. Pattern:recovery-guide requires errors to tell the LLM what to do next. Current errors are generic (e.g., 'Invalid start_cell format') with no suggested next steps or alternative actions.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 36 | 2026-07-28+ | v2 |
No input validation or constraint documentation. Parameters like 'cell' (A1, B2, etc.) and 'start_value'/'end_value' lack format/range constraints in descriptions. LLMs cannot infer valid formats and will pass invalid values.
Destructive operations (delete_sheet) lack confirmation or dry-run support. Pattern:confirmation-request recommends a confirm step for irreversible operations to prevent agent mistakes.