Loading ToolBench
Model Context Protocol server for DeepSeek's advanced language models.
Two tools with basic schemas and descriptions, but significant gaps in parameter documentation and error handling. chat_completion and multi_turn_chat both have input schemas visible in code (Zod definitions), but descriptions are minimal (12-18 chars). Parameters lack actionable guidance on constraints, formats, and dependencies. No output schema documented. Error handling is generic (try/catch with console.error). Tool names are verb-noun but lack clarity on when to use each (both are chat tools with overlapping functionality). No tool annotations (readOnlyHint, destructiveHint, idempotentHint). Resources are present but not integrated with tools.
Chat completion tool
Multi-turn chat tool
Tool descriptions are too short (12-18 chars). 'Chat completion tool' and 'Multi-turn chat tool' do not explain WHEN to use each tool, what distinguishes them, or what they return. LLMs cannot reliably select between them.
No output schema documented. Tools return text content but LLMs need to know the exact response structure (fields, types, nesting) to plan downstream calls and extract data reliably.
Parameter descriptions lack actionable constraints. 'Controls randomness in the output (0.0 to 2.0)' is present but other params like 'model' have no description at all. LLMs cannot infer valid values or when to use each parameter.
Error handling is generic (console.error with no recovery guidance). When API calls fail, the LLM receives no actionable error message, no indication of whether to retry, what went wrong, or how to fix it.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 46 | 2026-07-28+ | v2 |
Tool overlap not addressed. Both chat_completion and multi_turn_chat perform chat operations. The distinction (reasoner vs chat model, single message vs array) is buried in parameter names, not explained in descriptions. LLMs will struggle to choose correctly.