Loading ToolBench
Universal Multimodal Audio/Video Reading & Transcription MCP Server for AI Agents
OmniMedia defines 3 tools with complete input schemas and descriptions. Tool names follow verb_noun convention (inspect_media, read_audio, read_media). Descriptions are present and moderately detailed (80-150 chars), exceeding the 20-char minimum but falling short of the 50-200 char LLM-optimized range. All parameters have type definitions and descriptions. However, output schemas are not explicitly documented, tools return Union[Audio, str] or str without structured field definitions. Error handling uses ToolError for anticipated failures (ConfigError, ProviderRequestError, FileNotFoundError, ValueError), which is good, but recovery guidance is minimal. Tool annotations (readOnlyHint, destructiveHint, idempotentHint, openWorldHint) are present on all tools, demonstrating spec awareness. The code shows thoughtful parameter validation (e.g., _resolve_slice_start, _resolve_slice_minutes) and clear error messages, but output documentation is the main gap.
纯本地毫秒级探测音视频文件的时长、轨道、编码并给出多模态上下文预估（零网络、零凭证）。
直接读取本地音视频文件的音频流，返回原生音频数据或切片文件供当前宿主对话模型直接聆听。
调用外部模型端点（OpenAI、Gemini 等）对音视频进行转录、摘要、问答或自定义任务处理。
Output schemas not documented. Tools return Union[Audio, str] or str without specifying structured fields, field types, or what the LLM should expect. LLMs cannot plan downstream calls or extract data reliably.
Descriptions are functional but not LLM-optimized. At 80-150 chars, they lack explicit WHEN/WHY guidance. E.g., read_media description does not clarify when to use it vs read_audio, or what 'custom' mode requires.
read_media 'prompt' parameter is required only in 'custom' mode but not marked as conditional. Description says 'custom 模式必填' (required in custom mode) in Chinese, but schema does not enforce this dependency. LLMs may omit it in custom mode or pass it in other modes.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 59 | 2026-07-28+ | v2 |
Error messages are in Chinese. While culturally appropriate, LLMs trained primarily on English may misinterpret or fail to parse recovery guidance. Consider bilingual or English-primary error messages for agent compatibility.
No pagination or result-limiting guidance. read_media with 'summarize' or 'qa' modes could return very long transcripts or summaries. No documented limit or chunking strategy for context window management.