Humanizer: Power Platform MCP connector for detecting AI-sounding text

January 23, 2026

What is Humanizer?

Humanizer is an MCP-compliant Power Platform custom connector (Copilot Studio compatible) that detects 24 AI writing patterns across content, language, style, communication, and filler, based on Wikipedia’s “Signs of AI writing” guide. It exposes tools like humanize, detect_patterns, and get_patterns, returns an AI score (0–100), rewrite guidelines, and optional Application Insights telemetry.

Treat findings as prompts to revise, not absolute blockers. The goal is a clear, human tone.

Why this matters

• Use Humanizer’s 24 AI-writing patterns to keep content grounded
• Replace vague claims with specific facts, sources, and plain language
• Publish posts that read like you wrote them

Setup (Copilot Studio)

1. Import the Humanizer custom connector into your Power Platform environment using Humanizer/apiProperties.json and Humanizer/script.csx from the repo.
2. In Copilot Studio, go to Tools → Add a tool → Model Context Protocol → Invoke Humanizer MCP.
3. Create a connection (no auth required by default; add App Insights connection string in script.csx if you want telemetry).

Usage (Copilot Studio)

Agent instruction

Before responding, call the humanize action with the draft text. If aiScore > 25, revise using rewriteGuidelines and respond with the improved text.

Call humanize (full analysis + guidelines)

{
  "jsonrpc": "2.0",
  "method": "tools/call",
  "id": 1,
  "params": {
    "name": "humanize",
    "arguments": {
      "text": "Sample draft text for analysis...",
      "mode": "full",
      "preserveTone": true
    }
  }
}

Call detect_patterns (fast check by category)

{
  "jsonrpc": "2.0",
  "method": "tools/call",
  "id": 1,
  "params": {
    "name": "detect_patterns",
    "arguments": {
      "text": "Sample draft text for analysis...",
      "category": "Communication"
    }
  }
}

Call get_patterns (educate the user/agent)

{
  "jsonrpc": "2.0",
  "method": "tools/call",
  "id": 1,
  "params": {
    "name": "get_patterns",
    "arguments": {
      "category": "Language",
      "includeExamples": true
    }
  }
}

Sample response (humanize)

{
  "success": true,
  "aiScore": 42,
  "aiScoreDescription": "Moderate - Several AI patterns present",
  "patternsDetected": 3,
  "detectedPatterns": [
    {
      "patternId": "chatbot_artifacts",
      "patternName": "Chatbot Artifacts",
      "category": "Communication",
      "matchedText": "sample text",
      "suggestion": "Remove chatbot phrases entirely."
    },
    {
      "patternId": "significance_inflation",
      "patternName": "Significance Inflation",
      "category": "Content",
      "matchedText": "sample text",
      "suggestion": "State facts directly."
    }
  ],
  "rewriteGuidelines": "## Rewrite Guidelines\n- Remove chatbot artifacts...\n- Replace vague claims with specific facts...\n"
}

Customize patterns

Edit AIPatterns inside Humanizer/script.csx to add or adjust patterns (24 provided). Deploy the connector after updates.

Copilot Studio tips

• Use detect_patterns for quick checks; use humanize when you need guidelines.
• Keep preserveTone = true to retain formal/technical voice; set false to allow aggressive rewrites.
• Show get_patterns content to users to explain why their text was flagged.

Resources

• Humanizer connector: SharingIsCaring/Humanizer
• Tools: humanize, detect_patterns, get_patterns; Prompts: humanize_text, quick_check, rewrite_as_human, match_voice
• Wikipedia: Signs of AI writing