Loading ToolBench
A unified connector platform that exposes actions from multiple providers (including MCP servers, APIs, and custom integrations) through a standardized HTTP API and MCP interface. Supports credential management, action execution, OAuth flows, and catalog browsing.
Mixed quality across 16 tools. Dokploy server tools (1-12) have well-structured schemas with proper types and descriptions, but descriptions are formulaic ('Read/Modify Dokploy data via GET/POST') and lack actionable context for LLM selection. XYDC tools (13-16) have richer descriptions with warnings about costs and timeouts, but parameter descriptions are sparse. No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite clear READ_ONLY vs WRITE distinctions. Output schemas are not documented for any tool. Error handling guidance is absent.
Generate a category insight resource for a user-confirmed marketplace and category. XYDC waits up to five minutes and rebuilds failed resources internally. Each independent successful call costs 500 Credits, including existing-resource hits. Reuse resourceId for subsequent queries; do not blindly retry. Connector waits at most 55 seconds for generation. A timeout does not confirm upstream cancellation or prevent charges; the result may be unknown. Do not automatically retry generation after a timeout. Read xydc_mcp.read_category_insight_guide before the first category insight task.
Get daily newly observed advertising campaigns for one ASIN. This does not establish whether existing campaigns stopped, decreased, were deleted or remain active. Use traffic trends for traffic changes.
Get daily BSR category-ranking trends for one ASIN. Supports US, CA, MX, BR, UK, DE, FR, ES, IT and JP; this is not keyword ranking.
Get product titles, prices, currencies, star ratings, rating counts, images and Amazon links for up to 100 ASINs sharing one marketplace.
Formulaic descriptions lack LLM-actionable context. Dokploy tools use generic 'Read/Modify Dokploy data via GET/POST' patterns (12 tools). LLMs cannot distinguish when to call server-one vs server-all vs server-withSSHKey without reading schemas. Descriptions should explain the semantic difference and use case.
No output schemas documented. Tools return structured data but LLMs have no schema to understand response fields, forcing them to infer structure from examples or fail on unexpected formats. All 16 tools lack outputSchema declarations.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 68 | 2025-06-18+ | v2 |
Read Dokploy data via GET /server.all.
Read Dokploy data via GET /server.allForPermissions.
Read Dokploy data via GET /server.buildServers.
Read Dokploy data via GET /server.count.
Modify Dokploy state via POST /server.create.
Read Dokploy data via GET /server.getDefaultCommand.
Read Dokploy data via GET /server.one.
Read Dokploy data via GET /server.security.
Modify Dokploy state via POST /server.setup.
Modify Dokploy state via POST /server.setupMonitoring.
Read Dokploy data via GET /server.validate.
Read Dokploy data via GET /server.withSSHKey.
No tool annotations despite clear semantic distinctions. All Dokploy server-* tools marked READ_ONLY or WRITE in metadata, but no readOnlyHint or destructiveHint in tool definitions. XYDC tools lack idempotentHint despite cost warnings suggesting non-idempotent behavior.
XYDC tools lack parameter descriptions for optional fields. generate_category_insight_resource has intent_summary and user_task marked optional but descriptions are vague ('Optional purpose', 'Optional original user request'). Redaction guidance is buried in description text, not formalized as constraints.
No error handling guidance. Tools provide no recovery hints for common failures (e.g., invalid serverId, XYDC timeout after 55 seconds, cost exhaustion). LLMs cannot distinguish retryable vs fatal errors or know what to do next.