Loading ToolBench
Wrap a call to any remote LLM model and expose it as an MCP server tool to allow your main model to communicate with other models.
Two tools with minimal, under-documented schemas and descriptions. Both tools lack parameter descriptions, output schemas are undocumented, and error handling guidance is absent. The llm_call tool exposes a dynamic model parameter with no constraints, and ask_online_question has no meaningful schema depth. Parameter validation rules, enum constraints, and actionable error messages are missing. This server exhibits major definition gaps typical of early-stage implementations.
Asks an online question using the configured LLM.
Make a generic call to the configured LLM with a given prompt.
No parameter descriptions for 'model' (llm_call) and minimal description for 'prompt' (both tools). LLMs cannot infer parameter semantics without explicit text.
No output schemas documented for either tool. LLM cannot plan chaining calls, extract results, or validate responses.
llm_call exposes 'model' parameter with no enum, validation, or description. LLM will hallucinate invalid model names or reuse example values literally.
Error handling in ask_online_question_server.py returns generic error codes (-32000, -32602) with minimal guidance. LLM cannot self-correct or know whether to retry.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 33 | <=2025-11-25 | v2 |
| 2026-03-09 | F | 33 | 2024-11-05+ | v1 |
Tool descriptions lack WHEN to use, prerequisites, and idempotency statements. llm_call and ask_online_question appear to overlap (both call LLMs) but distinction is unclear.
No validation or constraints on dynamic 'model' parameter in llm_call. Placeholder '{max_user_prompt_tokens}' in 'prompt' description is not substituted, suggests incomplete template expansion.
LLMClient is instantiated with 'system_prompt_path' but tool descriptions do not mention that a system prompt is loaded. LLM has no visibility into how responses are shaped by system context.