MCP server for extracting and analyzing YouTube video transcripts with fact-checking capabilities
YouTube Transcript MCP has clear, action-verb naming (get_*, list_*, search_*, extract_*, find_*) and reasonable descriptions (avg ~100 chars). All 7 tools have input schemas with typed parameters and descriptions. However, output schemas are not documented in the source code, critical for LLM planning. Parameter descriptions lack format constraints (e.g., timestamp format 'MM:SS or HH:MM:SS' is stated but not enforced). Error handling is minimal; no recovery guidance or categorization. No tool annotations (readOnlyHint, etc.). The search_for_claim_verification tool lacks context on when to use it vs. other tools. Overall structure is functional but lacks production-grade polish.
Extract a specific segment of a transcript around a timestamp.
Find a specific claim in a transcript and return its timestamp and context.
Get chapter markers for a YouTube video.
Get the transcript for a YouTube video with timestamps in ~10 second intervals.
Get metadata for a YouTube video.
List available transcript languages for a YouTube video.
Search for information to help verify a claim made in a video.
Output schemas not documented. LLMs cannot plan downstream calls or extract required fields (e.g., does get_transcript return timestamps? structured segments?). Critical for tool composition.
No error handling guidance. Tools lack recovery hints (e.g., 'Video not found. Try a different URL or video ID.'). Agents cannot self-correct on failures.
Parameter format constraints stated in descriptions but not enforced in schema. 'timestamp' in extract_transcript_segment says 'MM:SS or HH:MM:SS' but no regex pattern or enum. LLMs may pass invalid formats.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 66 | <=2025-11-25 | v2 |
search_for_claim_verification lacks context on when to use it vs. other tools. Description does not explain prerequisites (e.g., must extract claim from transcript first) or relationship to find_claim_in_transcript.
No tool annotations (readOnlyHint, idempotentHint). All tools are read-only but this is not declared in schema, forcing LLMs to infer safety from descriptions.