Loading ToolBench
OpenAI-compatible LLM gateway and policy engine for AI agents, tool calling, audit logs, human approval, and MCP or LangChain integrations.
Single tool 'chat_completions' has a generic name that does not follow verb_noun conventions (should be 'forward_chat_completions' or 'evaluate_chat_completions'). Description is present but minimal (89 chars, below the 194-char baseline). Input schema visible but incomplete: 'messages' parameter lacks item schema definition (no 'items' key specifying message object structure). No output schema documented. No error handling guidance. Tool appears to be a gateway/proxy rather than a discrete agent action, which is architecturally misaligned with MCP's tool composition model.
Guard and forward chat completions requests with policy evaluation at input, tool, and output stages.
Tool name 'chat_completions' is not action-oriented. Does not start with a clear verb (get_, create_, forward_, evaluate_). LLMs cannot infer intent from the name alone.
Input schema for 'messages' parameter is incomplete. Declares type 'array' but lacks 'items' schema defining the structure of message objects (role, content fields). LLMs cannot validate message format.
No output schema documented. Tool description does not specify what the response contains (e.g., completion text, token counts, policy evaluation results). LLMs cannot plan downstream actions.
No error handling guidance. Tool description mentions 'policy evaluation' but does not explain what happens on policy violation, how to recover, or what error codes/messages to expect.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 35 | 2026-07-28+ | v2 |
Parameter descriptions missing. 'model' and 'messages' parameters lack descriptions explaining what values are valid, what format messages must follow, or what the model identifier should be.