aitells-mcp

MCP server for aitells — detect AI fingerprints in your text and rewrite in your voice, directly from Claude Code, Cursor, or any MCP-compatible client.

What is this

Two MCP tools:

• detect_ai_tells — paste any text, get a 0-100 humanness score plus exact positions of every AI fingerprint (em-dashes, "delve", parallel bullets, "in conclusion", and 9 more patterns). Free, no API key.
• humanize_text — rewrite AI-generated text matching the voice of writing samples you provide. Free first use, $19 lifetime unlock at aitells.vercel.app/rewrite.

Built after my own Reddit account got 2 public "all comments are AI generated" callouts in 24 hours. The patterns are findable; this tool just makes them visible (and fixable) before you hit post.

Install

npm install -g @perufitlife/aitells-mcp

Configure (Claude Code)

Edit ~/.claude.json (or your project's .claude.json):

{
  "mcpServers": {
    "aitells": {
      "command": "npx",
      "args": ["@perufitlife/aitells-mcp"]
    }
  }
}

For unlimited humanize_text after buying the $19 lifetime unlock:

{
  "mcpServers": {
    "aitells": {
      "command": "npx",
      "args": ["@perufitlife/aitells-mcp"],
      "env": {
        "AITELLS_LICENSE_EMAIL": "[email protected]"
      }
    }
  }
}

Configure (Cursor)

Add to ~/.cursor/mcp.json:

{
  "mcpServers": {
    "aitells": { "command": "npx", "args": ["@perufitlife/aitells-mcp"] }
  }
}

Use it

After config + reload:

"Use aitells to detect AI tells in this LinkedIn post I drafted: [paste]"

"Humanize this Reddit comment with these samples of how I write: [paste comment + samples]"

The model will pick the right tool automatically.

Tools reference

detect_ai_tells

| param | type | required | description | |---|---|---|---| | text | string | yes | Text to scan (max 30,000 chars) |

Returns: humanness score (0-100), grade label, list of flagged rules with severity + match positions.

humanize_text

| param | type | required | description | |---|---|---|---| | text | string | yes | AI-generated text to rewrite (max 8,000 chars) | | samples | string | recommended | 1-3 samples of how you actually write |

Returns: rewritten text in your voice.

The 12 rules

| # | Pattern | Severity | |---|---|---| | 1 | em-dash (—) | high | | 2 | "delve" | high | | 3 | "tapestry", "realm", "landscape", "journey", "endeavor" | high | | 4 | "navigate the X", "unlock the X", "harness the X" | medium | | 5 | "Great question", "Absolutely", "100%" as openers | high | | 6 | "In conclusion", "In summary", "Ultimately,", "At the end of the day" | high | | 7 | buzzword cluster (leverage, robust, seamless, holistic) | medium | | 8 | parallel bullet structure | medium | | 9 | tricolon rhythm (X, Y, and Z) | medium | | 10 | uniform sentence length | low | | 11 | Title Case headings in casual text | medium | | 12 | "However,", "Moreover,", "Furthermore," | low |

Pricing

• Detector: free forever, no API key, no rate limit beyond fair-use
• Humanizer: free first use per IP, then $19 lifetime at aitells.vercel.app/rewrite (first 100 buyers, then $49/mo)

Sister tools

• supabase-security — RLS auditor
• pocketbase-security, appwrite-security, firebase-security, nhost-security

License

MIT.