Server has 4 tools with basic descriptions and input schemas, but significant quality gaps. Tool names follow verb_noun convention (generate_form, create_google_form, list_user_forms, get_form_details), which is good. However, descriptions are minimal (10-50 chars), parameter descriptions are generic, and output schemas are completely undocumented. No error handling guidance, no pagination support for list_user_forms, and credentials_info parameter violates secret-injection pattern by exposing OAuth objects as parameters. Tools lack actionable error messages and recovery guidance.

• Move credentials_info from tool parameters to server-side session management. Retrieve credentials from the authenticated session context inside each tool, not from LLM-provided parameters. This prevents secrets from appearing in logs and traces.
• Document output schemas for all tools. For create_google_form, specify: {form_id: string, form_url: string, created_at: ISO8601}. For list_user_forms, specify: {forms: [{id, title, created_time, modified_time}], total_count: int}. For get_form_details, specify all returned fields with types.
• Expand parameter descriptions to 50-200 characters with specific constraints. E.g., 'form_schema: JSON object with required keys: title (string, 1-200 chars), description (string, 0-1000 chars), fields (array of field objects with type, label, required).'
• Add pagination to list_user_forms: document max_results behavior (default 20, max 100), add next_page_token or offset support, and return total_count so LLMs know if more results exist.
• Add error handling with recovery guidance. E.g., 'If authentication fails, ensure credentials are valid and have forms.body scope. If form not found, call list_user_forms() to verify the form_id.'

Inferred effective spec: <=2025-11-25.

• Relies on Logging (deprecated) - log to stderr or use OpenTelemetry