This MCP server exposes Google Sheets operations via three tools. While tool names are action-oriented (get_*, set_*) and input schemas are present with Zod validation, the definitions have critical gaps in completeness and clarity. All three tools use the `__gas_fakes` CLI wrapper to invoke Google Apps Script, which is a novel integration pattern but adds complexity. Tool descriptions are present but generic (10-60 chars), and output schemas are entirely undocumented, the agent cannot see what these tools return. Parameter descriptions exist for inputs but lack format guidance, ranges, or constraints. No error handling patterns are visible; the `__gas_fakes` wrapper catches exceptions and returns text output, but the agent receives no guidance on recovery or error classification. The tools follow basic verb_noun naming but lack the granularity to support agent-driven workflows (e.g., no batch operations, no pagination). Security review shows no secrets are exposed as parameters (credentials are handled by the `gas-fakes` CLI), which is positive.

• Document output schemas in server code. For get_sheet_names, return: { sheetNames: string[], count: number }. For get_values_from_sheet, return: { values: any[][], range: string, rowCount: number, colCount: number }. For set_values_to_sheet, return: { range: string, rowsAffected: number, columnsAffected: number }. This enables agents to extract and chain results.
• Expand all tool descriptions to 100-200 characters with WHEN, WHAT, and WHY. Example: 'Retrieves the list of all sheet names in a Google Spreadsheet. Call this first if you do not know sheet names before calling get_values_from_sheet. Requires a valid Spreadsheet ID.'
• Add format constraints to parameters. For 'range', state: 'A1Notation format, e.g. A1:D10 or Sheet1!A1:C5. Omit if you want the entire data range.' For 'values', state: 'Nested array of numbers and strings, e.g. [[1, 'a'], [2, 'b']]. Max 1000 rows, 26 columns.'
• Implement error recovery guidance. When sheet not found, return: { isError: true, content: [...], suggestion: 'Sheet not found. Available sheets: [list]. Did you mean one of these?' }. When invalid range, return: { isError: true, hint: 'Invalid A1Notation. Format: A1, A1:B2, or Sheet1!A1:C10.' }
• Add a dry_run parameter to set_values_to_sheet (optional, defaults to false). When true, validate the write without persisting, return what would be written, allowing the agent to confirm before committing.
• Consider a batch_get_values tool accepting { spreadsheetId, requests: [{sheetName, range?}] } and returning all values in one call, reducing sequential round-trips.