Loading ToolBench
Premiere MCP 剪辑助手，面向 Claude Code、Codex 和 OpenClaw，可结合文档、参考视频、提示词和本地素材目录完成视频剪辑规划与执行，当前更适合粗剪与初版装配。
8 tools with visible schemas and descriptions, but significant quality gaps. Tool names follow verb_noun pattern (premiere_get_*, premiere_create_*, etc.), which is good. However, descriptions are minimal (10-50 chars), parameter descriptions are sparse or missing, and output schemas are not documented. critic_edit_result and compare_result_to_blueprint have optional parameters with unclear semantics (blueprint vs editingBlueprintPath). No error handling guidance, no recovery patterns, no idempotency hints. Schemas use Zod with basic type constraints but lack detailed parameter documentation. This is typical of early-stage domain-specific tools that prioritize API coverage over LLM usability.
Compare editing result to an editing blueprint.
Evaluate and critique an editing result against goals, scenarios, and success criteria.
Insert an imported media item into the active Premiere sequence timeline.
Create a new sequence in the active Premiere project, optionally deriving settings from an imported media item.
Send the active Premiere sequence to Adobe Media Encoder.
Return the active Premiere project and sequence summary.
Import one or more media files into the active Premiere project.
Minimal parameter descriptions. Most parameters lack context on expected format, range, or constraints. E.g., 'trackIndex' has no min/max guidance; 'scenario' enum values are cryptic (natural_language, docx_guided, reference_video, viral_style) with no explanation of when to use each.
No output schema documentation. Tools return BridgeResult but the structure is not visible in source. LLMs cannot plan downstream calls or extract required fields without knowing what each tool returns.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 49 | <=2025-11-25 | v2 |
Open a Premiere project document from a given path.
Ambiguous parameter relationships in critic_edit_result and compare_result_to_blueprint. Both 'blueprint' (object) and 'editingBlueprintPath' (string) are optional, unclear which takes precedence or if both can be provided. No guidance on mutual exclusivity.
No error handling or recovery guidance. Tools return BridgeResult but no documentation on error codes, retryability, or what the LLM should do if a call fails (e.g., 'project not found', should it search for projects first?).
Tool descriptions are too brief (10-50 chars). Baseline for A+ tools is 50-200 chars. Current descriptions lack context on WHEN to use each tool, WHAT it modifies, and WHAT the LLM should expect. E.g., 'Return the active Premiere project and sequence summary' does not explain what fields are returned or how to use them downstream.