Server has 6 well-named tools with clear verb-noun patterns (upload_script, validate_script, translate_script, start_conversation, get_conversation_status, cancel_conversation). All tools have descriptions (50-150 chars, within baseline 34-392 range). Input schemas are present with typed parameters and descriptions. Tool annotations (readOnlyHint, destructiveHint) are implemented via _annotations() helper. However, output schemas are not documented in the visible code, responses are inferred from result objects but not formally declared. Error handling exists (JobNotFound, ScriptPayloadError, VoiceCatalogError exceptions) but recovery guidance is minimal. Parameter descriptions are adequate but lack format constraints (e.g., BCP-47 language tag format not validated in schema). No pagination or result limits documented for list-like operations.

• Document output schemas for all tools. Add a 'returns' field to each tool definition showing the structure of ScriptUploadResult, ScriptValidationResult, etc., with field types and descriptions.
• Add enum constraint for translate_script's target_language parameter. Define valid BCP-47 codes (e.g., 'de-DE', 'fr-FR', 'es-ES') as an enum or pattern to prevent invalid language codes.
• Enhance error messages with recovery guidance. When JobNotFound is raised, return 'Job {job_id} not found. Verify the job_id from start_conversation or check if it has expired.' instead of a bare exception.
• Implement a dry-run mode for start_conversation. Add an optional 'dry_run' boolean parameter (default false) that validates the script and returns estimated cost without initiating synthesis.
• Document polling best practices in get_conversation_status description. Specify recommended delay (e.g., 5-10 seconds), max retries (e.g., 120 for 10-20 min timeout), and backoff strategy.