Three read-only tools with clear naming (list_sheets, get_sheet, get_recent_sheets) and solid descriptions (avg 180 chars). All tools have input schemas with type definitions and parameter descriptions. However, output schemas are not formally documented in the code, only described in docstrings. Error handling is present but minimal (ValueError for missing sheets, no recovery guidance). Parameter descriptions are good but lack explicit constraints (e.g., max_rows bounds, regex pattern format). No tool annotations (readOnlyHint, idempotentHint) despite all being read-only. Composition is clean, each tool has one responsibility and tools chain well (list_sheets → get_sheet). Security is solid: read-only scope enforced, no secrets in params, OAuth handled server-side.

• Add formal output schemas using TypedDict or Pydantic models with field descriptions. Document return types in tool signatures so LLMs see structured expectations.
• Annotate all three tools with readOnlyHint=true and idempotentHint=true in FastMCP tool decorators to signal safety to agents.
• Enhance error messages with actionable recovery steps. E.g., 'Sheet "meeting-notes" not found. Call list_sheets(spreadsheet_id) to see available sheets, or check the sheet name spelling.'
• Add explicit parameter constraints: max_rows as integer with minimum=1, maximum=100000; pattern as string with description 'Valid Python regex (e.g., ^\\d{6}$)'; count as integer with minimum=1, maximum=50.
• Implement pagination for get_sheet: add offset parameter (default 0), document that results are capped at 1000 rows, return row_count and total_rows so agents know if more data exists.