Naturalmelo

Lightweight MCP server for AI content detection — detects AI-template patterns in text.

One tool only:

check_ai_content

How it works

1. Rule-based scan (always runs, no API key needed) — 5 built-in pattern rules covering generic openings, hollow claims, filler transitions, stacked connectors, and marketing hype.
2. LLM scan (optional, BYOK) — if OPENAI_API_KEY is set, a second pass with GPT-4o-mini catches subtler patterns and merges results with the rule scan.

What it returns

• patternScore (0–100) — higher = more AI-template patterns
• naturalnessLabel — human-readable tier ("Mostly natural" → "Strong AI writing signals")
• hits — list of flagged issues, each with:
  • id, label, severity (warn / fail)
  • message — what's wrong
  • excerpt — the exact text flagged
  • humanizeHint — how to rewrite it
• language, wordCount, createdAt

Quick Start

npm install
npm run build

Works out of the box with zero configuration (rule-based scan only). Add OPENAI_API_KEY for better accuracy.

Environment variables

Variable	Required	Default	Description
OPENAI_API_KEY	No	—	Your OpenAI API key. Without it, only rule-based scan runs.
OPENAI_BASE_URL	No	https://api.openai.com/v1	Custom OpenAI-compatible endpoint.
OPENAI_MODEL	No	gpt-4o-mini	Model override.

MCP Config Example

{
  "mcpServers": {
    "naturalmelo-mcp": {
      "command": "node",
      "args": ["<ABSOLUTE_PATH>/naturalmelo-mcp/dist/index.js"],
      "env": {
        "OPENAI_API_KEY": "sk-your-key-here"
      }
    }
  }
}

Tool Input

{
  "text": "In today's rapidly evolving digital landscape, AI has become essential...",
  "mode": "general"
}

• text (required) — minimum 20 characters
• mode (optional) — "general" (default) or "academic"

Notes

• Rule-based scan runs locally — no network calls, no API keys.
• LLM scan uses your own OpenAI key (BYOK) — you control usage and billing.
• Academic mode skips connector-stacking and hype-flagging rules.
• Never commit .env or API keys into git.