MCP server for YouTube research providing tools to search videos, retrieve metadata, fetch transcripts, and access channel information
Server has 5 well-named tools with clear verb-noun patterns (youtube_search, youtube_video_info, youtube_transcript, youtube_channel_info, youtube_channel_videos). All tools have descriptions (avg ~150 chars, within baseline 34-392 range). Input schemas are present with type definitions and parameter descriptions. However, output schemas are NOT documented, tool descriptions state what fields are returned (e.g., 'id, title, description, duration, view_count, published_at, channel') but no formal schema is provided. Error handling returns strings instead of structured objects, making it hard for LLMs to parse failures. No tool annotations (readOnlyHint, destructiveHint, idempotentHint). Parameter descriptions are adequate but lack explicit constraints (e.g., max_results max=50 is mentioned in description but not enforced in schema). Overall: solid naming and descriptions, but missing output schemas and error structure.
Get channel metadata by URL, handle, or ID. Returns: id, title, description, subscriber_count, view_count, video_count, published_at, thumbnail, custom_url.
List recent videos from a channel. Accepts channel URL, @handle, or ID. Returns videos sorted by date (newest first) with metadata: id, title, description, duration, view_count, published_at.
Search YouTube videos by query. Returns a list of videos with metadata: id, title, description, duration, view_count, published_at, channel.
Get subtitles/transcript for a YouTube video. Accepts a video URL or ID. Returns timestamped text. Looks for Russian subtitles first, then English by default.
Get video metadata by URL or ID. Returns: id, title, description, duration, view_count, published_at, channel. Direct lookup by video ID — no search required.
Output schemas not documented. Tool descriptions mention returned fields (id, title, duration, etc.) but no formal JSON Schema is provided. LLMs cannot plan downstream tool calls or extract fields reliably without knowing the response structure.
Error responses are unstructured strings instead of typed objects. Tools return 'YouTube API error 403: ...' or 'Video not found: ...' as plain strings. LLMs cannot distinguish error types (retryable vs. user-fixable vs. fatal) or extract recovery guidance.
No tool annotations. Tools lack readOnlyHint (all are read-only), destructiveHint, or idempotentHint. This prevents agents from understanding which tools are safe to retry and which have side effects.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 59 | 2026-07-28+ | v2 |
Parameter constraints not enforced in schema. max_results description says 'max 50' but schema has no maximum constraint. lang parameter accepts array but no enum or validation. LLMs may pass invalid values.
youtube_transcript returns plain text instead of structured object. Response is a formatted string with timestamps. Should return array of {timestamp, text} objects so LLMs can filter, search, or summarize transcripts programmatically.