Ezkielyna MCP

An MCP server that lets your AI assistant ask other AI models for a second opinion, compare 2-4 answers side by side, search the web and check your balance, all through the Ezkielyna API. Works in Claude Code, Claude Desktop, Cursor, OpenCode and any other MCP client that runs local stdio servers.

Example (in Claude Code): "Ask deepseek-v4.1-flash and glm-5.3 to review the SQL in db/report.sql for injection risks, then tell me where they disagree."

🇮🇩 Bahasa Indonesia: lihat bagian di bawah.

• One file, no dependencies: only Node.js 18 or newer. No npm install.
• Answers come back with tokens used, latency and an estimated cost at list price.
• Paid calls are capped per session, max_tokens is capped, and the API key is never printed.

Requirements

1. Node.js 18+ (node -v). Get the LTS from nodejs.org.
2. An Ezkielyna API key in group Standard (DeepSeek, GLM, Kimi, MiniMax, Qwen and other text models) or Auto (also Claude, GPT and Grok at the Premium 1.5x multiplier). Buy API Credit at ezkielyna.store/beli-api-key (from $0.5, QRIS/DANA/GoPay/USDT), redeem it under Wallet at api.ezkielyna.store, then create a key under API Keys.
3. list_models works without a key (it shows the public price list), so you can look before you buy.

Tools

Tool	What it does	Cost
ask_model	Sends a prompt (optional system) to one model. model default deepseek-v4.1-flash, max_tokens default 1024 (cap 4096), optional temperature. Returns the answer, tokens, latency and estimated cost.	paid, 1 call
compare_models	Same prompt to 2-4 models in parallel; answers side by side with tokens, latency and cost each. A failing model is reported without hiding the others.	paid, 1 call per model
list_models	Models your key can use (GET /v1/models) joined with public prices (USD per 1M tokens, or per call). Optional filter, e.g. deepseek, claude, tts.	free
web_search	Web search through POST /v1/search: query, max_results 1-10 (default 5), optional time_range (day/week/month/year) and language.	counts toward the cap
check_balance	Credit limit, usage and remaining balance reported by the OpenAI-compatible billing endpoints, plus calls left this session. If the endpoint answers in an unexpected shape, it points you to the dashboard instead of guessing.	free

The other models cannot see your conversation: the assistant must put the code or context into the prompt. Prices shown are list prices (USD) from the public price list and include the lowest group multiplier the model is in; actual bills can be lower thanks to cache hits. Exact charges are in Usage Logs on api.ezkielyna.store.

Price example (public price list, 11 Nov 2026): deepseek-v4.1-flash costs $0.04 per 1M input tokens and $0.16 per 1M output tokens, so a 2,000-token prompt with a 1,000-token answer is about $0.00024.

Install

With npx (recommended): nothing to install. The configs below run npx -y ezkielyna-mcp, which downloads the package (one file) on first start.

From source:

git clone https://github.com/EzkielynaGroup/ezkielyna-mcp.git
cd ezkielyna-mcp
node --test test/        # optional: local mock server, no network, no cost

Then use "command": "node" with "args": ["/full/path/to/ezkielyna-mcp.mjs"] instead of npx.

Windows: if the app cannot start npx, use "command": "cmd" with "args": ["/c", "npx", "-y", "ezkielyna-mcp"]. Write source paths in JSON with forward slashes (C:/Users/you/ezkielyna-mcp/ezkielyna-mcp.mjs).

Add it to your app

Replace sk-... with your key.

Claude Code

claude mcp add ezkielyna -e EZKIELYNA_API_KEY=sk-... -- npx -y ezkielyna-mcp

Add --scope user to make it available in every project. Check with claude mcp list.

Claude Desktop: Settings → Developer → Edit Config (claude_desktop_config.json), then quit and reopen Claude Desktop.

{
  "mcpServers": {
    "ezkielyna": {
      "command": "npx",
      "args": ["-y", "ezkielyna-mcp"],
      "env": { "EZKIELYNA_API_KEY": "sk-..." }
    }
  }
}

Cursor: ~/.cursor/mcp.json (all projects) or .cursor/mcp.json (one project), same JSON as Claude Desktop.

OpenCode: opencode.json in the project, or ~/.config/opencode/opencode.json.

{
  "$schema": "https://opencode.ai/config.json",
  "mcp": {
    "ezkielyna": {
      "type": "local",
      "command": ["npx", "-y", "ezkielyna-mcp"],
      "enabled": true,
      "environment": { "EZKIELYNA_API_KEY": "sk-..." }
    }
  }
}

Configuration (environment)

Variable	Required	Default	Notes
EZKIELYNA_API_KEY	yes (except list_models)		key in group Standard or Auto
EZKIELYNA_BASE_URL	no	https://api.ezkielyna.store/v1	API address
EZKIELYNA_MAX_CALLS	no	50	calls per session for ask_model, compare_models (one per model) and web_search. A guard against runaway agent loops; restart the app to reset.
EZKIELYNA_MAX_TOKENS	no	4096	upper limit for max_tokens (larger requests are lowered to it)
EZKIELYNA_TIMEOUT_MS	no	120000	request timeout (search uses at most 30 s)

Safety and limits

• The key is only sent in the Authorization header to EZKIELYNA_BASE_URL; it is scrubbed from every answer and log line. Logs go to stderr; stdout carries only MCP messages.
• Session cap (EZKIELYNA_MAX_CALLS) is checked before anything is sent; compare_models reserves one slot per model. Calls the gateway clearly refused (wrong key, no balance, busy) give their slot back.
• Prompts are limited to 100,000 characters; answers longer than 60,000 characters (20,000 per model in compare_models) are truncated.
• No files are read or written. Every request has a timeout.
• Model ids containing uncensored, abliterated or nsfw are never listed and cannot be called through this server.

Data source

Prices come at runtime from the public endpoint https://api.ezkielyna.store/api/pricing (no key, cached 10 minutes). Price per 1M tokens = model_ratio × 2 × group multiplier, output price = input price × completion_ratio (the New API pricing convention), which matches the published list at https://api.ezkielyna.store/llms-full.txt (checked 11 Nov 2026). Nothing is bundled.

FAQ

How do I get a second opinion from another AI model inside Claude Code? Add this server (claude mcp add ezkielyna ...) and ask, for example, "compare deepseek-v4.1-flash and glm-5.3 on this function". Claude calls compare_models and shows both answers with their cost.

Can I use DeepSeek or GLM inside Claude Desktop or Cursor? Yes. ask_model sends a prompt to any model your key can use; run list_models to see the ids and prices.

How much does one question cost? It depends on the model and length; each answer shows the estimated cost. With deepseek-v4.1-flash a typical code-review question (a few thousand tokens) costs a fraction of a US cent.

Does it send my whole conversation to the other model? No. Only the prompt and optional system text the assistant passes to the tool are sent.

Can it search the web? Yes, web_search returns up to 10 results with title, URL and snippet through the Ezkielyna search endpoint.

Is Ezkielyna affiliated with DeepSeek, Anthropic or OpenAI? No. Ezkielyna is an independent API gateway and is not affiliated with the model vendors or the MCP client makers.

Development

node --test test/

The tests spawn the server over stdio against a local mock gateway: no internet, no cost. Issues and pull requests are welcome.

---

Bahasa Indonesia

Server MCP agar asisten AI Anda (Claude Code, Claude Desktop, Cursor, OpenCode) bisa bertanya ke model AI lain untuk pendapat kedua, membandingkan 2-4 jawaban berdampingan, mencari di web, dan mengecek saldo lewat Ezkielyna API.

Contoh (di Claude Code): "Minta deepseek-v4.1-flash dan glm-5.3 me-review SQL di db/laporan.sql untuk risiko injection, lalu jelaskan di mana mereka berbeda pendapat."

Yang dibutuhkan

1. Node.js 18+. Cukup satu file, tanpa npm install.
2. API key Ezkielyna grup Standard (DeepSeek, GLM, Kimi, MiniMax, Qwen, dll.) atau Auto (termasuk Claude, GPT, Grok dengan pengali Premium 1.5x). Beli API Credit mulai $0.5 lewat QRIS/DANA/GoPay/USDT di ezkielyna.store/beli-api-key, tukarkan di menu Wallet di api.ezkielyna.store, lalu buat key di menu API Keys.
3. list_models bisa dipakai tanpa key (menampilkan daftar harga publik).

Tool

Tool	Fungsi	Biaya
ask_model	Kirim prompt (opsional system) ke satu model. Default deepseek-v4.1-flash, max_tokens 1024 (maks 4096). Hasil: jawaban, token, waktu, perkiraan biaya.	berbayar, 1 panggilan
compare_models	Prompt yang sama ke 2-4 model sekaligus (paralel), jawaban berdampingan dengan token, waktu, dan biaya masing-masing.	berbayar, 1 panggilan per model
list_models	Model yang bisa dipakai key Anda + harga publik. Bisa difilter, mis. deepseek.	gratis
web_search	Pencarian web (1-10 hasil), opsional time_range dan language (mis. id).	dihitung ke batas sesi
check_balance	Batas kredit, pemakaian, dan sisa saldo dari endpoint billing gaya OpenAI. Jika format jawabannya tidak dikenal, Anda diarahkan ke dashboard.	gratis

Model lain tidak melihat percakapan Anda: asisten harus menyertakan kode/konteks di dalam prompt. Harga yang ditampilkan adalah harga daftar (USD); tagihan sebenarnya ada di Usage Logs dashboard.

Pemasangan

Claude Code:

claude mcp add ezkielyna -e EZKIELYNA_API_KEY=sk-... -- npx -y ezkielyna-mcp

Claude Desktop, Cursor dan OpenCode: pakai contoh JSON di bagian Inggris di atas, cukup ganti sk-.... Di Windows, jika npx tidak jalan, pakai "command": "cmd" dan "args": ["/c", "npx", "-y", "ezkielyna-mcp"].

Pengaturan dan keamanan

• EZKIELYNA_MAX_CALLS (default 50): batas panggilan per sesi untuk ask_model, compare_models (dihitung per model) dan web_search, dicek sebelum permintaan dikirim. Mulai ulang aplikasi untuk mereset.
• EZKIELYNA_MAX_TOKENS (default 4096): batas atas max_tokens.
• Key tidak pernah ditampilkan di jawaban maupun log. Tidak ada file yang dibaca atau ditulis.
• Model dengan id berisi uncensored, abliterated atau nsfw tidak ditampilkan dan tidak bisa dipanggil.

Tanya jawab

Bagaimana cara minta pendapat kedua dari model AI lain di Claude Code? Pasang server ini lalu minta, misalnya, "bandingkan jawaban deepseek-v4.1-flash dan glm-5.3 untuk fungsi ini". Claude akan memanggil compare_models.

Berapa biaya satu pertanyaan? Tergantung model dan panjang teks; setiap jawaban menampilkan perkiraan biaya. Dengan deepseek-v4.1-flash, satu pertanyaan review kode biasanya jauh di bawah 1 sen dolar.

Apakah Ezkielyna berafiliasi dengan DeepSeek, Anthropic atau OpenAI? Tidak. Ezkielyna adalah API gateway independen.

License

MIT, Copyright (c) 2026 Ezekiel Lieberhart (Ezkielyna). Ezkielyna is an independent API gateway and is not affiliated with the model vendors or the MCP client makers.