Transcribe any podcast episode — CLI + MCP server. Three tiers: Gemini, OpenAI, or free local Whisper.
Server has 4 tools with complete schemas and descriptions. Naming follows verb_noun convention (transcribe_*, search_*, detect_*). Descriptions are adequate (100-200 chars) but lack actionable guidance on when to use each tool vs. alternatives. Parameter descriptions are present but generic. Output is unstructured text strings rather than typed objects, forcing LLMs to parse results. No error recovery guidance. Missing tool annotations (readOnlyHint, idempotentHint). Tier parameter enum is well-defined; output_format enum is good. Overall solid foundation but lacks LLM-optimization patterns.
Check which transcription tier is currently available. Returns the detected tier and available options.
Search for podcasts by name or topic.
Transcribe a local audio file.
Transcribe a podcast episode from a URL. Accepts Apple Podcasts URLs, Spotify URLs, RSS feed URLs, show name + episode number, or direct audio URLs.
All tools return unstructured text strings instead of typed JSON objects. LLMs must parse free-form text, wasting tokens and increasing hallucination risk. E.g., transcribe_podcast returns a single string; should return {text: string, format: string, duration_seconds?: number, language?: string}.
Tool descriptions lack actionable guidance on when to use each tool. E.g., transcribe_podcast and transcribe_audio both transcribe, the description should clarify: 'Use transcribe_podcast for URLs/feeds; use transcribe_audio for local files.' This prevents LLM confusion and wasted calls.
No tool annotations (readOnlyHint, idempotentHint). All tools are read-only and idempotent, but LLMs cannot infer this. Add tool annotations to signal safety and enable aggressive caching/retry strategies.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 56 | 2026-07-28+ | v2 |
Error responses are generic strings ('Error: {e}') with no recovery guidance. E.g., 'Error: Could not resolve URL' should suggest 'Try search_podcast_tool first to find the feed URL.' This guides LLM recovery without extra reasoning.
search_podcast_tool returns unstructured text with feed URLs and Apple URLs mixed in. Should return a structured array: [{show_name: string, feed_url: string, apple_url?: string, description?: string}]. This enables downstream tool chaining and reduces parsing errors.