MCP server for interacting with Google Drive and Google Sheets
Server has 5 tools with mixed quality. Tool definitions exist but lack rigor: descriptions are present but generic (avg ~60 chars, below 194-char baseline); schemas are partially visible but incomplete; parameter descriptions are sparse or missing; no output schema documentation; error handling is minimal. The 'greet' tool is a placeholder with no business value. gdrive_search and gsheets_read have reasonable parameter coverage but lack constraint documentation (e.g., pageSize max=100 stated in description but not in schema). gsheets_update_cell and gdrive_read_file are minimal. No tool annotations (readOnlyHint, destructiveHint, idempotentHint). Composition is reasonable (search→read→update chain possible) but output schemas are not documented, forcing LLMs to guess what fields are returned.
Read contents of a file from Google Drive
Search for files in Google Drive
Greet a person with a welcome message to Google Drive
Read data from a Google Spreadsheet with flexible options for ranges and formatting
Update a cell value in a Google Spreadsheet
Output schemas not documented. LLMs cannot infer what fields gdrive_search, gdrive_read_file, gsheets_read, and gsheets_update_cell return. This breaks tool chaining and forces agents to guess field names.
Parameter descriptions are minimal or missing. 'pageToken' in gdrive_search has a description, but 'pageSize' lacks constraint details (max=100 is in tool description, not schema). 'range' in gsheets_update_cell lacks examples of valid A1 notation. LLMs cannot reliably format inputs.
No tool annotations (readOnlyHint, destructiveHint, idempotentHint). gsheets_update_cell is marked WRITE in metadata but the tool definition lacks destructiveHint. Agents cannot determine retry safety without explicit hints.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 47 | 2026-07-28+ | v2 |
'greet' tool is a placeholder with no business value. It clutters the tool list and wastes agent reasoning cycles. Remove or replace with a meaningful tool.
No error handling guidance. Tools do not return actionable error messages. E.g., if gdrive_search fails due to invalid query syntax, the agent receives a raw error with no recovery hint (e.g., 'Try simplifying your query' or 'Use quotes for exact matches').