# JevK5

JevK5-4B (Qwen3.5-4B + distilled LoRA) reads a text and typed questions, then chooses one of the answers you listed for each question. CPU only. The same call returns the decisions plus usage and latency.

**Service id:** `jevk5`  
**Version:** `0.2.0`  
**Status:** `available`

**Canonical payloads:** [https://mcp.glc-rag.hu/guide/jevk5/payload](https://mcp.glc-rag.hu/guide/jevk5/payload) · [markdown](https://mcp.glc-rag.hu/guide/jevk5/payload.md) · `docs://jevk5/payload`

## Authentication

MCP endpoint: `https://mcp.glc-rag.hu/mcp` (streamable HTTP)

**Agents (recommended):** self-register with `account_type=agent` to get an
auto-approved token — see [https://mcp.glc-rag.hu/guide/agent](https://mcp.glc-rag.hu/guide/agent).

Or register as a human on the public site (all listed services are auto-approved),
wait for system-admin approval, then create a token.

```http
Authorization: Bearer mcp_...
```

Cursor `mcp.json` example:

```json
{
  "mcpServers": {
    "jevk5": {
      "url": "https://mcp.glc-rag.hu/mcp",
      "headers": {
        "Authorization": "Bearer mcp_YOUR_TOKEN"
      }
    }
  }
}
```

## Tools

### `jevk5_status`

Whether JevK5-4B is loaded, plus the CPU thread budget. Free — 0 credits.

**Input schema:**

```json
{
  "type": "object",
  "properties": {},
  "additionalProperties": false
}
```

**Examples:**

```json
{}
```

### `jevk5_decide` — Decide

Pick one allowed answer per question with JevK5-4B. Pass `text` and `questions` (each with `id`, `question`, and `options`). Returns answers plus usage and latency. 1 credit per 10 questions.

**Input schema:**

```json
{
  "type": "object",
  "additionalProperties": false,
  "required": [
    "text",
    "questions"
  ],
  "properties": {
    "text": {
      "type": "string",
      "minLength": 1,
      "maxLength": 60000,
      "description": "Text the questions are about"
    },
    "questions": {
      "type": "array",
      "minItems": 1,
      "maxItems": 32,
      "description": "Questions and the answers the model may pick",
      "items": {
        "type": "object",
        "additionalProperties": false,
        "required": [
          "id",
          "question",
          "options"
        ],
        "properties": {
          "id": {
            "type": "string",
            "minLength": 1,
            "maxLength": 64
          },
          "question": {
            "type": "string",
            "minLength": 1,
            "maxLength": 500
          },
          "options": {
            "type": "array",
            "minItems": 2,
            "maxItems": 32,
            "uniqueItems": true,
            "items": {
              "type": "string",
              "minLength": 1,
              "maxLength": 200
            }
          }
        }
      }
    }
  }
}
```

**Examples:**

```json
{
  "text": "J\u00f3 napot! A m\u00falt heti rendel\u00e9sem (R-1842) m\u00e9g mindig nem \u00e9rkezett meg, holnapra viszont m\u00e1r kellene. K\u00e9rem, n\u00e9zz\u00e9k meg, mi t\u00f6rt\u00e9nt, \u00e9s ha lehet, k\u00fcldj\u00e9k ki m\u00e9g ma.",
  "questions": [
    {
      "id": "tema",
      "question": "Mir\u0151l sz\u00f3l a lev\u00e9l?",
      "options": [
        "sz\u00e1ll\u00edt\u00e1s",
        "sz\u00e1mla",
        "\u00faj aj\u00e1nlat",
        "egy\u00e9b"
      ]
    },
    {
      "id": "surgosseg",
      "question": "Mennyire s\u00fcrg\u0151s?",
      "options": [
        "r\u00e1\u00e9r",
        "hamarosan",
        "ma kell"
      ]
    }
  ]
}
```

## Usage notes

Call `jevk5_decide` with `text` and `questions`. Each question has `id`, `question`, and `options` (the allowed answers). The response `answers` list has one row per question: `answer` is one of that question's options, `scores` is the probability of each option, `confidence` is the model's confidence in the pick.

Also returned: `usage.input_tokens` / `usage.output_tokens` (always 0), `latency_ms` (model time), `client_wall_ms` (full tool wall time).

Billing: 1 credit per 10 questions, rounded up (1–10 costs 1, 11–20 costs 2). `jevk5_status` is free. A failed call is not billed.

Backend: `alibiserikbay/JevK5` on CPU. Context up to 16_384 tokens. At most 32 questions and 32 options per question (options beyond 16 use knockout passes). Prefer clear option labels (e.g. `igaz`/`hamis`, not `IGAZ`/`HAMIS`).

## Errors / limits

Missing text, duplicate ids or options, or a model failure → `{error, is_error}`. Those calls cost 0 credits.

## Agent discovery

- Agent registration: `https://mcp.glc-rag.hu/guide/agent`
- Markdown: `https://mcp.glc-rag.hu/guide/jevk5.md`
- Index: `https://mcp.glc-rag.hu/llms.txt`
- MCP resource: `docs://jevk5`