Loading ToolBench
MCP server for AI video production pipeline
32 tools with highly inconsistent quality. Most tools have basic descriptions (10-50 chars) that lack actionable context for LLM selection. Parameter schemas are present but sparse: many lack descriptions, enums, or constraints. No documented output schemas visible. Error handling and recovery guidance absent. Tool names follow verb_noun convention (positive), but descriptions fail to explain WHEN to use each tool or what distinguishes similar tools (e.g., generate-veo-prompt vs breakdown-shot). No evidence of idempotency markers, permission gates, or security patterns. Composition is reasonable (single responsibility per tool), but chaining is fragile, output fields not documented, forcing LLMs to guess what data flows between tools.
Add text overlay to video
Analyze audio file for speech, music, and ambient sounds
Analyze song using librosa for audio features
Analyze video with Gemini 2.5 Flash
Assemble video from shots with transitions, overlays, and audio layers
Break a shot into takes with Veo prompts and transition strategies
Browse and search AI-generated video clips from Google Veo
Create a new project
Descriptions are uniformly terse (20-40 chars). None explain WHEN to use the tool, what distinguishes it from similar tools, or what the LLM should expect in return. E.g., 'Get project details' lacks context on whether this returns structure, status, or metadata.
No output schemas documented. LLMs cannot plan downstream tool calls or extract required fields (e.g., project_id, job_id) without seeing what each tool returns. This forces agents to guess or make wasteful discovery calls.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 41 | 2026-07-28+ | v2 |
Detect faces in video using face_recognition library
Execute project: generates Veo prompts and submits jobs for all shots
Generate first/last frame image prompts from a Veo prompt
Generate image via Vertex AI Imagen 3
Generate music using ElevenLabs
Generate sound effect using ElevenLabs
Generate shot structure from concept using Claude
Generate text-to-speech audio using ElevenLabs
Generate Veo video prompt from action description
Get detailed analysis of a video clip
Get metadata for a video clip
Get current configuration
Get job status and result
Get project details
List available API endpoints for the video generator service
List all background jobs
List all projects
Lock a project to prevent concurrent modifications
Search clips by tag
Set configuration values
Trim video to specified start and end times
Unlock a project
Upscale video to higher resolution
Validate project structure and content
Parameter descriptions missing or minimal. E.g., 'arc' in generate-structure lists enum values but does not explain what each arc type produces or when to choose one. 'style' in generate-veo-prompt has no description at all.
No error handling or recovery guidance. Tools do not document what can fail, how to detect failure, or what the LLM should do next. E.g., generate-image may fail if Vertex AI quota is exceeded, but no guidance is provided.
No tool annotations (readOnlyHint, destructiveHint, idempotentHint). LLMs cannot distinguish safe read-only tools from destructive ones (e.g., lock-project vs get-project). This risks unintended side effects.