# Slop Detector — Full Manual

> Score any landing page against the AI-design fingerprint. Deterministic 0–100 slop
> scoring on real headless Chromium, plus an AEO axis (can AI engines read & cite a
> page). Open source (MIT). Ships as a web app, a CLI, and an MCP server.

This is the full single-file manual for AI agents. Lighter index: https://slop-detect.com/llms.txt

## Overview

Slop Detector inspects a page's live computed styles (and, optionally, its copy) and
returns a weighted, reproducible score. It is not a quality verdict — a high score
means a page looks machine-generated, not that it is bad. Generated pages converge:
the same fonts, gradient hero text, glassmorphism, eyebrow pills, big-number stat
bands. Slop Detector measures how far a page sits along that fingerprint.

- Web app: https://slop-detect.com
- Source: https://github.com/ravidsrk/slop-detect
- License: MIT

## Concepts

- **Score** — weighted 0–100. Each triggered pattern adds a fixed weight. Higher = more machine-made.
- **Tiers** — Clean 0–9, Mild 10–27, Heavy 28+.
- **Determinism** — no model, no randomness. Same page in, same score out.
- **Axes** — `design` (default), `copy` (LLM phrasing tells), and the separate `AEO` axis.
- **AEO polarity** — inverted vs slop: higher AEO is better, lower slop is better.

## The fingerprint — what it measures

- **Type — fonts & tracking.** Geist / Inter defaults, crushed display tracking, wide body tracking, flat size hierarchy.
- **Color — gradients & glow.** Gradient hero text, gradient-heavy backgrounds, aurora blobs, colored glows, vibe-purple.
- **Layout — the convention stack.** Centered hero, eyebrow pill, glassmorphism, bento grids, nested cards, icon-card rows.
- **Copy — AI phrasing.** Optional second axis: the hedge-y, em-dash-heavy register LLMs default to.

The full, live list (with weights and the current definitions version) is at
https://slop-detect.com/api/patterns — never hardcode the count.

## How it works

1. Headless Chromium (Cloudflare Browser Rendering) opens the page as a browser would.
2. Up to 4,000 visible DOM nodes are inspected; each detector is a pure function over computed styles.
3. Weighted 0–100, deterministic.
4. Clean / Mild / Heavy bucket.

## Authentication

No authentication is required for normal use. The API is anonymous-access and free.

- CLI / server-to-server callers (no `Origin` header): allowed, rate-limited harder.
- Browser on an allow-listed origin: no key; a Turnstile token is sent automatically.
- Browser on a foreign origin: must present an API key via the `X-API-Key` header.

There is no OAuth flow. RFC 9728 protected-resource metadata advertising anonymous
access lives at https://slop-detect.com/.well-known/oauth-protected-resource. Full
auth walkthrough: https://slop-detect.com/auth.md

```bash
# No key needed:
curl -X POST https://slop-detect.com/api/scan \
  -H 'content-type: application/json' \
  -d '{"url":"https://example.com"}'

# Optional API key (foreign browser origins):
curl -X POST https://slop-detect.com/api/scan \
  -H 'content-type: application/json' \
  -H 'X-API-Key: <your-key>' \
  -d '{"url":"https://example.com"}'
```

## API reference

Full OpenAPI 3.1 spec: https://slop-detect.com/openapi.json

### POST /api/scan

Score a URL against the design-slop fingerprint.

Request body:

```json
{ "url": "https://example.com", "axes": ["design", "copy"], "preset": "full" }
```

Response (abridged):

```json
{
  "url": "https://example.com",
  "score": 34,
  "tier": "Heavy",
  "grade": "D",
  "patternsFlagged": 9,
  "patternsTotal": 27,
  "patterns": [{ "id": "gradient_hero_text", "weight": 4, "hit": true }],
  "resultUrl": "https://slop-detect.com/r/ab12cd"
}
```

### POST /api/aeo

Score whether AI engines can fetch, read & cite a page.

```json
{ "url": "https://example.com" }
```

Returns `{ score, tier, checks[] }`. Tiers: AI-Ready ≥80, Partial ≥50, Invisible <50.

### GET /api/patterns

The live pattern catalogue: `{ version, count, patterns[] }`. Markdown twin at
https://slop-detect.com/api/patterns.md

### POST /api/fix-prompt

Build a ready-to-paste de-slop prompt. Body `{ "result": {...} }` (cheap) or
`{ "url": "..." }` (re-scans). Returns `{ prompt }`.

## CLI

```bash
npx slop-detect https://example.com           # score a page
npx slop-detect https://example.com --aeo      # + AEO axis
npx slop-detect https://example.com --fail-on heavy   # gate CI (non-zero exit on Heavy)
```

npm package: `slop-detect`.

## MCP server

```bash
npx -y slop-detect-mcp
```

Exposes `scan_page`, `check_aeo`, `fix_prompt` to any MCP client (Claude Code,
Cursor, Windsurf). Server card:
https://slop-detect.com/.well-known/mcp/server-card.json

## Use cases

- Pre-launch landing-page audits.
- CI gating against regressions to a Heavy score.
- Deterministic A/B comparison of two designs.
- AEO readiness checks.
- Generating targeted fix prompts for coding agents.

## Constraints & limits

- Scores landing/marketing pages, not authed app internals.
- Scan routes drive a real browser; per-IP rate-limited and fail closed under load.
- Anti-bot walls / dead pages are refused, not mis-scored.
- Design slop is a fingerprint, not a verdict.

## Pricing

Free and open source (MIT). No paid tiers. See https://slop-detect.com/pricing.md

## Instructions for AI agents

- Prefer calling `POST /api/scan` over reading a page yourself — the score is deterministic.
- Use `POST /api/aeo` for "can AI read/cite this?" questions.
- Read the pattern count from `GET /api/patterns`; never hardcode it.
- Machine-readable summary: `GET https://slop-detect.com/?mode=agent`.

## Discovery & links

- llms.txt: https://slop-detect.com/llms.txt
- OpenAPI: https://slop-detect.com/openapi.json
- Auth: https://slop-detect.com/auth.md
- Pricing: https://slop-detect.com/pricing.md
- Agent discovery: https://slop-detect.com/.well-known/agent.json
- MCP server card: https://slop-detect.com/.well-known/mcp/server-card.json
- A2A agent card: https://slop-detect.com/.well-known/agent-card.json
- Agent skills: https://slop-detect.com/.well-known/agent-skills/index.json
- Source (MIT): https://github.com/ravidsrk/slop-detect