Loading ToolBench
An MCP server that bridges the gap between your applications and various LLM providers
MindBridge has 3 tools with complete input schemas and descriptions. getSecondOpinion is well-parameterized with 13 inputs covering LLM configuration (temperature, top_p, reasoning_effort, etc.), but descriptions are minimal (10-50 chars) and lack guidance on when/why to use each parameter. listProviders and listReasoningModels have empty input schemas (correct for discovery tools) but minimal descriptions. No output schemas are documented, LLMs cannot predict response structure. Error handling returns text but lacks recovery guidance. Tool names follow verb_noun pattern (good), but parameter descriptions are sparse and don't explain constraints or dependencies.
Get responses from various LLM providers
List all configured LLM providers and their available models
List all available models that support reasoning capabilities
No output schemas documented. LLMs cannot predict response structure for getSecondOpinion (returns content array?), listProviders (object with models array?), or listReasoningModels. Forces LLMs to guess field names and types.
Parameter descriptions are too brief (10-50 chars). 'Temperature parameter for response randomness' lacks guidance on typical values (0.7 for balanced, 0.2 for deterministic). 'Reasoning effort level for models that support it' doesn't explain low/medium/high trade-offs or which models support it.
Error handling returns plain text but lacks recovery guidance. 'Provider not found' error lists available providers (good), but 'Model not found' error doesn't suggest how to discover valid models. No distinction between retryable vs fatal errors.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 49 | 2026-07-28+ | v2 |
Parameter dependencies undocumented. reasoning_effort only applies to certain models/providers, but this is only mentioned in a console.warn(), not in the parameter description. LLMs won't see the warning and may pass invalid combinations.
Discovery tools (listProviders, listReasoningModels) lack descriptions explaining when to call them. Should state: 'Call this first to see available providers before invoking getSecondOpinion' to guide multi-step planning.