AppSheet MCP has solid naming conventions (verb_noun pattern) and complete input schemas with type definitions. However, descriptions are generic and lack LLM-optimization guidance. Parameter descriptions are minimal (10-30 chars), missing format constraints, dependencies, and recovery hints. Output schemas are undocumented, LLMs cannot infer what fields to expect from responses. Error handling is absent; no guidance on retryability, user-fixable vs fatal errors, or recovery paths. Security is well-handled (no secrets in params, server-side injection). Tool composition is good (single responsibility, chainable IDs), but descriptions don't explain WHEN to use each tool or dependencies between them.

• Document output schemas for all tools. E.g., appsheet_find_rows should specify: 'Returns array of row objects. Each row is a dict mapping column names (strings) to values (string|number|boolean|null). Example: [{"Name": "Alice", "Status": "Active", "Age": 30}]'
• Expand parameter descriptions to 50-100 chars with format constraints. E.g., selector: 'AppSheet filter expression using column names in brackets, e.g. [Status]="Active" AND [Age]>25. Supports =, <>, <, >, <=, >=, AND, OR operators.'
• Add WHEN-to-use guidance in tool descriptions. E.g., appsheet_describe_table: 'Call this first to discover available columns and their types before using add_rows or update_rows. Returns sample row with all column names.'
• Implement error recovery hints in tool descriptions. E.g., appsheet_find_rows: 'If table not found, call appsheet_list_tables to see available tables. If selector syntax fails, check column names with appsheet_describe_table.'
• Add dry-run support to appsheet_delete_rows. Accept optional 'dry_run=true' parameter that returns which rows would be deleted without actually deleting them, preventing accidental data loss.