Loading ToolBench
Local vision MCP server for DeepSeek/Claude — image description + video frame analysis + audio transcription via Ollama + MiniCPM-V 4.6 + faster-whisper
Three tools with Chinese descriptions and basic schemas. All tools have input schemas with required parameters and type definitions. However, descriptions are minimal (20-150 chars) and lack LLM-optimized guidance on WHEN to use each tool or WHAT to expect. No output schemas documented. Parameters lack descriptions in most cases. No error handling guidance. Tool names are action verbs (describe_*) but descriptions don't explain dependencies or prerequisites. No pagination, no structured output guidance, no recovery paths.
提取视频/音频文件的音频轨道，分析音频元数据（时长、编码、采样率等）。支持本地文件路径或 HTTP/HTTPS URL。
用本地视觉模型（MiniCPM-V 4.6）描述图片内容。传入图片文件的绝对路径。
分析视频内容：自动提取关键帧，逐帧用视觉模型描述，返回完整时间线总结。支持本地文件路径或 HTTP/HTTPS URL。分析完成后自动清理临时文件。
No output schemas documented. LLMs cannot plan downstream operations or extract structured data from responses.
Parameter 'path' lacks description. LLMs cannot infer whether to pass absolute paths, relative paths, or URLs without explicit guidance.
Tool descriptions are minimal (20-150 chars) and lack WHEN/WHY guidance. 'Describe image content' does not explain when to call describe_image vs describe_video vs describe_audio, or what format the response takes.
No error handling guidance. If Ollama is unreachable, ffmpeg fails, or a file is invalid, LLMs receive raw errors with no recovery path.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 47 | 2026-07-28+ | v2 |
describe_video 'interval' parameter lacks description. LLMs cannot determine valid range, default behavior, or impact on output quality.