MCP server that forwards prompts to an OpenAI-compatible LLM endpoint and returns responses
Single tool 'ask_llm' has a basic schema with one required parameter (prompt: string), but lacks critical production-grade quality markers. Description is minimal (67 chars, below 10-char baseline for context). No parameter description visible in schema, only type. No output schema documented. No error handling guidance. No tool annotations (readOnlyHint, destructiveHint, idempotentHint). Tool name lacks action verb clarity ('ask_llm' is vague; 'send_prompt_to_llm' or 'query_llm' would be clearer). No pagination, no result limits, no structured output specification. Naming does not follow verb_noun pattern strongly enough.
Forwards a prompt to an OpenAI-compatible LLM endpoint and returns the response
Tool description is too short (67 chars) and lacks context. Should explain WHAT it does, WHEN to use it, and what it returns. Current: 'Forwards a prompt to an OpenAI-compatible LLM endpoint and returns the response', no guidance on output format, error cases, or prerequisites.
Input parameter 'prompt' has no description in the schema. LLMs cannot infer whether this expects a single sentence, multi-line text, system instructions, or a specific format. Add description: 'The prompt text to send to the LLM (string, any length; will be sent as-is to the endpoint).'
No output schema documented. LLMs do not know what fields to expect in the response (e.g., is it a string, an object with 'text' and 'tokens', or an array?). Document the return type explicitly.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 32 | <=2025-11-25 | v2 |
No error handling guidance. If the LLM endpoint is unreachable, rate-limited, or returns an error, the tool provides no recovery hints. Add error classification: retryable (network timeout), user-fixable (invalid endpoint), or fatal (auth failure).
Tool name 'ask_llm' is vague and does not follow verb_noun convention strongly. 'query_llm', 'send_prompt_to_llm', or 'call_llm_endpoint' would be clearer and help LLMs infer intent from the name alone.