slopCheck

MCP server that detects AI slop in your content — fabricated claims, false uniqueness, histrionic language, and other substance issues that AI writing tools introduce.

Existing "AI detectors" catch writing style. slopCheck catches substance: is what AI wrote actually true?

What it detects

10 categories, ~20 patterns:

| Category | Examples | Severity | |----------|----------|----------| | Fabricated Claims | "saves 500 hours", "73% of developers" | High | | False Uniqueness | "the only tool that", "nobody else built" | High | | Histrionic Language | "game-changer", "revolutionary", "killer feature" | Medium | | Phantom Citations | "studies show", "according to experts" | High | | Inflated Credentials | "taught 500+ students" (actual: 185) | Medium | | Misleading Completeness | "all tests pass", "fully tested" | Medium | | Generic Defaults | "[INSERT]", "Acme Corp", template testimonials | Low | | AI Formatting Artifacts | Emoji section headers (the ChatGPT tell) | Low | | Fabricated Narrative | "I was using Cursor", "My Mac was at 14GB" | High | | Wrong Audience | Technical jargon in marketing copy (LLM mode) | Low |

Two modes

• No API key: Regex-only detection (~20 patterns). Works fully offline.
• With ANTHROPIC_API_KEY: Adds Haiku-powered claim analysis for deeper substance checking.

Personal slop profile

The build_profile tool mines your Claude Code session history for correction patterns — finding where you've called out the AI for making things up. This builds a weighted profile that boosts detection sensitivity for YOUR specific weak spots.

Install

npm install -g @iteachyouai/slopcheck

Claude Code MCP config

Add to ~/.claude.json:

{
  "mcpServers": {
    "slopcheck": {
      "command": "npx",
      "args": ["-y", "@iteachyouai/slopcheck"],
      "env": {
        "ANTHROPIC_API_KEY": "your-key-here"
      }
    }
  }
}

The ANTHROPIC_API_KEY is optional — without it, slopCheck runs in regex-only mode.

MCP Tools

scan_content

Scan text for slop. Returns flagged items with category, severity, confidence, and suggestions.

Input: { content: "your text here", severity_threshold?: "low"|"medium"|"high" }
Output: Score (0-100), flags with suggestions

check_claim

Analyze a specific claim for verifiability.

Input: { claim: "saves 500 hours per year", context?: "optional surrounding text" }
Output: Issues found, verification checklist

list_patterns

List all detection patterns with descriptions and examples.

Input: { category?: "fabricated-claim" }
Output: Pattern library with examples

build_profile

Mine your session history to build a personal slop profile.

Input: {}
Output: Profile summary, writes ~/.slopcheck/profile.json

Programmatic use

import { scanContent } from "@iteachyouai/slopcheck";

const result = scanContent("Studies show this game-changer saves 500 hours per year.");
console.log(result.score);  // 0-100
console.log(result.flags);  // Array of flagged items

Development

git clone https://github.com/ITeachYouAI/slopcheck.git
cd slopcheck
npm install
npm test
npm run test:coverage

Contributing

See CONTRIBUTING.md for guidelines on adding patterns, running tests, and submitting PRs.

License

MIT