aether-mcp

The Model Context Protocol is a JSON-RPC standard from Anthropic for letting AI clients call out to external tool servers. aether-mcp is one of those servers — when you add it to your AI client's config, the client sees a set of tools like aether_chat, aether_deobfuscate_js, etc., and can invoke them mid-conversation.

The user-facing effect: stay in Claude Desktop / Cursor / Cline / Zed for normal work, and when you ask a question the main AI declines to answer, it routes the same prompt through Aether and shows you the answer in the same chat tab. No new app to learn, no context-switch to a browser.

• Node 18 or newer (the MCP server runs as npx -y aether-mcp)
• An AI client with MCP support: Claude Desktop, Cursor, Cline, Zed, Continue, or any other
• An Aether API key — generate one at https://trynoguard.com/account. Keys start with ak_live_.

Every MCP client has the same essential config shape. Drop this snippet into the right file, restart the client, and the Aether tools appear. Replace ak_live_… with your real key.

Claude Desktop

Edit ~/Library/Application Support/Claude/claude_desktop_config.json on macOS, or %APPDATA%\Claude\claude_desktop_config.json on Windows.

json
{
  "mcpServers": {
    "aether": {
      "command": "npx",
      "args": ["-y", "aether-mcp"],
      "env": { "AETHER_API_KEY": "ak_live_..." }
    }
  }
}

Cursor

Edit ~/.cursor/mcp.json (global) or .cursor/mcp.json in a project (per-project).

json
{
  "mcpServers": {
    "aether": {
      "command": "npx",
      "args": ["-y", "aether-mcp"],
      "env": { "AETHER_API_KEY": "ak_live_..." }
    }
  }
}

Cline (VS Code extension)

In Cline's settings, click MCP Servers, then Edit MCP Settings. That opens cline_mcp_settings.json:

json
{
  "mcpServers": {
    "aether": {
      "command": "npx",
      "args": ["-y", "aether-mcp"],
      "env": { "AETHER_API_KEY": "ak_live_..." },
      "disabled": false,
      "autoApprove": []
    }
  }
}

Zed

Open Settings → Open Settings JSON and add this under context_servers:

json
{
  "context_servers": {
    "aether": {
      "command": {
        "path": "npx",
        "args": ["-y", "aether-mcp"],
        "env": { "AETHER_API_KEY": "ak_live_..." }
      }
    }
  }
}

Continue

Edit ~/.continue/config.json and add to the mcpServers array:

json
{
  "mcpServers": [
    {
      "name": "aether",
      "command": "npx",
      "args": ["-y", "aether-mcp"],
      "env": { "AETHER_API_KEY": "ak_live_..." }
    }
  ]
}

These appear in your AI client's tool panel after restart. Most are uncensored versions of the kind of analysis other AIs refuse — RE, deobfuscation, anti-cheat, malware analysis.

The killer use case: pair aether-mcp with ida-pro-mcp (mrexodia's open-source MCP server for IDA Pro's analysis API). Both run in your Claude Desktop / Cursor session simultaneously.

You ask things like:

Decompile sub_18000A4C0 from IDA, identify what kind
of integrity check it is, and suggest a bypass.

Your AI client orchestrates: ida-pro-mcp calls IDA to get the C output → passes it to aether_identify_pattern → passes the result to aether_suggest_bypass. Three steps, zero refusals, all in one chat message.

json
{
  "mcpServers": {
    "aether": {
      "command": "npx",
      "args": ["-y", "aether-mcp"],
      "env": { "AETHER_API_KEY": "ak_live_..." }
    },
    "ida-pro": {
      "command": "uv",
      "args": ["run", "ida-pro-mcp"]
    }
  }
}

Each tool call uses your Aether credits. Typical costs:

Plans start at $15/month for 2,500 credits, or top up $5 = 1,500 credits on the free tier. Top-up credits never expire.

• Per-key rate limit. FREE: 30/hr · STARTER: 200/hr · PRO: 1,000/hr · UNLIMITED: 100,000/hr. Sliding 60-min window, returned as response headers.
• Per-key IP fingerprinting. The MCP server runs locally and reaches our API over HTTPS. Each request is tagged with a SHA-256 hash of the source IP, so abuse from a single key across many machines is detectable.
• Credit reservation pattern. Each tool call reserves the worst-case cost up front, then refunds the unused portion. Concurrent fan-out attacks can't over-spend.
• Identity scrubbing. Server-side filter strips any leaked underlying-model brand names from responses before they reach the MCP server. Multi-language coverage.