Single tool with solid naming (verb_noun: get_transcript) and comprehensive description (287 chars, within 10-1024 baseline). Input schema is complete with three parameters, all typed and described. However, output schema is undocumented, the tool returns a string (Markdown with YAML frontmatter) but LLMs cannot infer the structure or fields to extract. Error handling is good (friendly_error function maps exceptions to actionable messages), but no recovery guidance in tool description. Missing tool annotations (readOnlyHint present via Risk field, but not in schema). No pagination or result limits documented despite potentially large transcript outputs.

• Document output schema in tool description: 'Returns Markdown string with YAML frontmatter containing: title, show, duration, language, source_url, sections (with timestamps if requested), and transcript text.'
• Add recovery guidance to description: 'If Spotify/Apple resolution fails, try the direct YouTube URL. Spotify-exclusive and Apple-exclusive episodes are not supported.'
• Add readOnlyHint annotation to input schema to signal safety: {"type": "object", "properties": {...}, "readOnlyHint": true}
• Document result limits: 'Transcripts are returned as-is; very long episodes (6+ hours) may exceed context limits. Consider splitting via include_timestamps=true to identify sections.'
• Constrain language parameter with enum or pattern: 'ISO 639-1 language code (e.g., en, es, fr, de). Defaults to en. Availability depends on YouTube caption availability.'

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

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