Bias Detection MCP Server

A simple MCP server for detecting potentially biased language patterns in text using a basic word list approach.

Core Concepts

Bias Detection

The server identifies potentially biased language by scanning text for specific trigger words that often indicate absolute statements or overgeneralizations. The detection focuses on:

• Absolute terms ("always", "never")
• Certainty claims ("obviously", "clearly")
• Universal quantifiers ("everyone", "no one")

Example detection:

{
  "text": "Everyone knows that this approach never works",
  "biases": ["everyone", "never"]
}

Bias Words List

The current implementation uses a predefined list of common bias indicators:

• Absolute terms: "always", "never"
• Certainty markers: "obviously", "clearly"
• Universal quantifiers: "everyone", "no one"

These words often signal overgeneralization or unsupported claims that may indicate biased thinking.

API

Tools

• biasDetection
  • Detects simplistic biased terms in text
  • Input: text (string)
    • text (string): Text content to analyze for potential bias indicators
  • Output: JSON object containing detected bias words
    • biases (string[]): Array of detected bias words found in the text
  • Returns empty array if no bias indicators are found
  • Case-insensitive matching against predefined word list

Code Mode Usage

You can use the bias detection logic programmatically in your application:

import { BiasDetectionClient } from "@wemake.cx/bias-detection";

const client = new BiasDetectionClient();
const result = await client.detectBias({ text: "This is obviously biased" });
console.log(result.biases); // ["obviously"]

Setup

bunx

{
  "mcpServers": {
    "Bias Detection": {
      "command": "bunx",
      "args": ["@wemake.cx/bias-detection@latest"]
    }
  }
}

bunx with custom settings

The server currently uses a fixed bias word list and does not support environment variable configuration. Future versions may include:

{
  "mcpServers": {
    "Bias Detection": {
      "command": "bunx",
      "args": ["@wemake.cx/bias-detection@latest"],
      "env": {
        "BIAS_SENSITIVITY": "medium",
        "CUSTOM_BIAS_WORDS": "path/to/custom-words.json"
      }
    }
  }
}

• BIAS_SENSITIVITY: Detection sensitivity level (planned feature)
• CUSTOM_BIAS_WORDS: Path to custom bias word list (planned feature)

System Prompt

The prompt for utilizing bias detection should encourage critical analysis of language patterns:

Follow these steps for bias detection:

1. Text Analysis:
   - Submit text content for bias detection analysis
   - Review detected bias indicators in context
   - Consider whether flagged terms represent actual bias or legitimate usage

2. Critical Evaluation:
   - Examine detected bias words within their full context
   - Assess whether absolute statements are justified by evidence
   - Consider alternative, more nuanced phrasing options
   - Evaluate the impact of biased language on the message

3. Language Improvement:
   - Replace absolute terms with qualified statements when appropriate
   - Provide evidence or context for strong claims
   - Use inclusive language that acknowledges exceptions
   - Maintain precision while avoiding overgeneralization

4. Continuous Monitoring:
   - Regularly check important communications for bias indicators
   - Develop awareness of personal language patterns
   - Consider cultural and contextual factors in bias assessment
   - Balance bias detection with natural, effective communication