Loading ToolBench
Autonomous Media Studio - MCP server for video/audio editing, media ingestion, timeline management, and rendering with hardware-accelerated FFmpeg
MCPelies defines 5 tools with complete JSON schemas and descriptions in Portuguese. All tools have proper input schemas with typed parameters and enums where appropriate (track_type, media_type, provider). However, descriptions lack LLM-optimization guidance (e.g., no 'when to use' context, no dependency hints, no recovery guidance). Parameter descriptions are minimal (avg ~40 chars vs baseline 72). No output schemas documented. Error handling is absent from tool definitions. Tools follow single-responsibility principle but lack composition guidance. Naming is clear (verb_noun pattern) but descriptions are domain-specific Portuguese without English fallback, limiting LLM understanding.
Adiciona um asset (vídeo, áudio ou overlay gráfico) à timeline de edição do projeto no SQLite.
Faz a ingestão de um arquivo de mídia local (vídeo/áudio), extrai metadados via ffprobe e salva no banco SQLite. Cria um projeto automaticamente se necessário.
Dispara o FFmpeg para juntar todos os assets da timeline e renderizar o MP4 final usando aceleração de hardware.
Busca vídeos ou áudios em bancos de imagem/som gratuitos baseando-se no texto, baixa o arquivo para o disco e retorna o asset_id.
Verifica se a timeline atual do projeto possui conflitos físicos (sobreposições na mesma track exclusiva).
No output schemas documented. LLMs cannot infer what fields to expect from tool responses, forcing them to guess at downstream field names and types. This breaks tool chaining and wastes tokens on clarification.
Descriptions lack LLM-optimization context. No 'when to use' guidance, no dependency hints (e.g., 'call ingest_media first'), no recovery guidance on failure. Descriptions are domain-specific Portuguese without English translation, limiting cross-language LLM understanding.
No error handling guidance in tool definitions. Errors like 'file not found' or 'invalid project_id' are handled in code but not documented. LLMs receive raw errors with no recovery path (e.g., 'Try search_and_download_stock() first').
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 53 | <=2025-11-25 | v2 |
Parameter descriptions are minimal (avg ~40 chars vs baseline 72). E.g., 'ID do projeto atual' lacks context on how to obtain project_id or what happens if it's invalid. No constraints on numeric ranges (start_in_timeline, duration could be negative).
No idempotency or confirmation pattern for destructive operations. render_project and add_to_timeline modify state but lack dry-run or confirmation steps. Agents cannot safely retry on ambiguous failures.