Loading ToolBench
🎬 The FIRST MCP server for Kling AI video generation! Generate, extend, lip-sync AI videos, virtual try-on, and manage your account directly from Claude.
Server has 5 tools with complete input schemas and descriptions. Naming follows verb_noun pattern (generate_*, check_*, extend_*, create_*). Descriptions are present and substantive (100-300 chars), exceeding the 10-char minimum. However, several parameters lack descriptions (e.g., camera_control.config properties have descriptions but nested structure could be clearer). Output schemas are not documented, LLMs cannot predict return structure. Error handling is absent from tool definitions. No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite clear WRITE/READ_ONLY risk classifications. Parameter constraints are well-defined (enums, min/max bounds), but some optional parameters lack clarity on defaults and side effects.
Check the status of a video generation task
Create a lip-sync video by synchronizing mouth movements with audio. Supports both text-to-speech (TTS) with various voice options or custom audio upload. The original video must contain a clear, steady human face with visible mouth. Works with real, 3D, or 2D human characters (not animals). Video length limited to 10 seconds.
Extend a video by 4-5 seconds using Kling AI. This feature allows you to continue a video beyond its original ending, generating new content that seamlessly follows from the last frame. Perfect for creating longer sequences or adding additional scenes to existing videos.
Generate a video from an image using Kling AI
Generate a video from text prompt using Kling AI
Output schemas not documented. LLMs cannot predict return structure (task_id, status, video_url, etc.). Forces agents to guess field names for downstream calls.
No tool annotations (readOnlyHint, destructiveHint, idempotentHint). Risk classifications exist (WRITE, READ_ONLY) but are not exposed in tool definitions. Agents cannot distinguish safe-to-retry operations from destructive ones.
No error handling guidance in tool descriptions. Agents have no recovery path if API fails (e.g., 'task_id not found', 'quota exceeded', 'invalid prompt'). Errors should guide LLM to retry, ask user, or call alternative tool.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 61 | 2026-07-28+ | v2 |
Nested parameter descriptions incomplete. camera_control.config properties (horizontal, vertical, pan, tilt, roll, zoom) have descriptions, but the parent camera_control object lacks clarity on when it applies (V2 models only) and mutual exclusivity with other parameters.
Default values not explicitly stated in descriptions. Parameters like model_name, aspect_ratio, duration, mode, cfg_scale have defaults mentioned in descriptions but not formalized. Agents may not recognize defaults and pass redundant parameters.