Loading ToolBench
MCP stdio server and CLI for letting desktop AI agents safely inspect and validate a local video production pipeline.
Fusion Cast Agent Kit has complete tool schemas and basic descriptions, but falls short on LLM-optimized guidance. All 5 tools have input schemas with type definitions and required fields properly declared. However, descriptions are minimal (10-20 chars) and lack context about WHEN to use each tool, what prerequisites exist, or what the output structure contains. Parameter descriptions are entirely absent, 'order' and 'workspace' parameters have no guidance on format, constraints, or dependencies. No output schemas are documented. Error handling is generic (unknown tool error only). The tool names follow verb_noun convention (fusion_cast_*) but are domain-specific and could be clearer about their operational impact (e.g., 'order_gate' doesn't signal it's a validation gate). No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite clear risk levels (READ_ONLY, WRITE, IRREVERSIBLE). Missing guidance on idempotency, retry safety, and recovery paths for destructive operations.
Read order metadata and latest local reports for one Fusion Cast commercial order.
Run the operation entry in validate mode, or execute mode when explicitly requested.
Run the commercial order input gate for one order.
Run the final commercial release gate for one order.
Detect the local Fusion Cast pipeline and list available commercial orders.
Parameter descriptions entirely missing. 'order', 'workspace', and 'execute' parameters have no guidance on format, constraints, or valid values. LLMs cannot infer whether 'order' is a UUID, slug, or numeric ID.
Tool descriptions are too brief (10-20 chars) and lack context. 'Detect the local Fusion Cast pipeline and list available commercial orders' is 68 chars but does not explain WHEN to call this vs inspect_order, what the output structure is, or what 'workspace' parameter controls.
No output schemas documented. Tools return JSON via textContent() but LLMs have no schema to understand what fields to expect. This breaks downstream tool chaining and forces LLMs to guess field names.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 42 | 2026-07-28+ | v2 |
No tool annotations despite clear risk levels. fusion_cast_order_gate and fusion_cast_release_gate are marked WRITE/IRREVERSIBLE but have no readOnlyHint, destructiveHint, or idempotentHint in the schema. Agents cannot distinguish safe from dangerous operations.
No error recovery guidance. callTool() throws generic 'Unknown tool' error. No guidance on retryability, user-fixable vs fatal errors, or what to do if an order is not found or validation fails.