Oriora model-router as an MCP server — ask for the best AI model per task; you run the call on your own keys (BYOK). Oriora never sees your prompt or vendor key.
Single tool 'recommend_model' has a clear, descriptive name and good description (189 chars, within 10-1024 baseline). Input schema is present with two parameters, both typed and described. However, output schema is completely undocumented, the tool returns JSON via text content but provides no structured schema definition for the response fields (model, alternatives). Parameter 'vendors' lacks type constraints (should be enum or pattern). No error handling guidance for common failures (invalid task_type, API errors, missing key). Tool is read-only and low-risk, but schema incompleteness and missing output documentation prevent higher scores.
Ask Oriora's router for the best AI model for a task. Returns a 'vendor/model' id + cheaper/faster alternatives. You run the call yourself on your own keys — Oriora never sees your prompt or vendor key. Costs $0.001 per recommendation on your Oriora account.
Output schema completely undocumented. Tool returns JSON with 'model' and 'alternatives' fields but provides no structured schema definition. LLMs cannot plan downstream operations or validate response structure.
Parameter 'vendors' is an array of strings with no enum, pattern, or format constraint. Description says 'restrict to vendors you hold keys for' but does not list valid vendor names (e.g., openai, anthropic, google). LLMs will hallucinate invalid vendor names.
Error handling provides no recovery guidance. When ORIORA_API_KEY is missing, response says 'Set ORIORA_API_KEY' but does not explain how to obtain one or what to do next. When API returns 4xx/5xx, raw JSON is returned without actionable next steps.
Parameter 'task_type' description provides examples ('coding, reasoning, knowledge, general') but does not declare them as an enum. LLMs may invent task types outside this set. Should use enum constraint instead of prose examples.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 59 | 2026-07-28+ | v2 |
STDIO transport only, hard cap at 50 for protocol readiness. Server is not remotely accessible and cannot be used by hosted MCP clients. Requires local stdio connection only.