Loading ToolBench
MCP server for ClipForge — drive the keyless AI video pipeline (create_video, generate_script, list_voices, get_video …) from any MCP client. Thin stdio wrapper over a running ClipForge instance; stock + Edge-TTS are keyless, only script generation needs an LLM key.
ClipForge MCP has 5 tools with mixed quality. Tool naming follows verb patterns (create_video, generate_script, list_voices, get_video, list_projects), but descriptions vary widely in quality and actionability. Parameter schemas are present and mostly complete with types and enums, but several tools lack depth in parameter descriptions and output documentation. The server lacks error handling guidance, recovery hints, and proper output schema documentation. Descriptions are generally brief and do not explain WHEN to use each tool, dependencies between tools, or failure modes. Output schemas are not explicitly documented in the tool definitions.
Create a finished commerce short video from product images and description, or from a topic sentence
Generate a video script using an LLM for either product commerce or topic narration
Retrieve the final video URL and metadata for a completed composition
List all projects in the ClipForge instance
List available text-to-speech voices (Edge TTS and custom voices)
Output schemas not documented. Tool definitions describe inputs but provide no explicit schema for responses. LLMs cannot plan downstream tool chains or extract required fields from results.
Missing error handling and recovery guidance. Tool descriptions do not indicate failure modes, retryable errors, or what the agent should do if a call fails. E.g., create_video does not document what happens if LLM config is invalid or if video composition fails.
Incomplete parameter descriptions. Many parameters lack detail on valid ranges, formats, or constraints. E.g., 'targetDuration' in create_video states it defaults to 25 seconds but does not specify minimum/maximum bounds or units clarity.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 55 | 2026-07-28+ | v2 |
Tool descriptions too brief and lack context. E.g., list_projects has only 'List all projects in the ClipForge instance' (45 chars). This does not explain when to call it, what structure it returns, or how it chains to other tools. Descriptions should be 50 - 200 chars with WHEN/WHY context.
Missing idempotency guidance. create_video and generate_script are stateful operations but lack documentation on whether repeated calls with identical input are safe to retry or produce duplicate side effects.
Dependency hints missing. The tools suggest a workflow (generate_script → create_video → get_video) but do not document this in tool descriptions. LLMs must infer the sequence without explicit guidance, leading to wasted calls and errors.
Parameter 'llmConfig' is complex object but lacks detailed schema. It expects 'baseUrl', 'apiKey', 'model' but the description does not explain which LLM providers are supported, whether apiKey should be masked/injected server-side (security concern), or format of baseUrl.
No pagination documentation. list_voices and list_projects may return unbounded results, but neither tool accepts limit/offset/page parameters or documents max result count. Large result sets will waste tokens and degrade LLM reasoning.