# Slop Detector

> Score any landing page against the AI-design fingerprint. Deterministic 0–100
> slop scoring for landing pages, plus an AEO axis that measures whether AI
> engines can read and cite a page. Open source; runs on real Chromium; ships as
> a web app, a CLI, and an MCP server.

Slop Detector inspects a page's live computed styles (and, optionally, its copy)
and returns a weighted, reproducible score. Lower is better: 0 means no
AI-design tells were found. It is not a quality verdict — a high score means a
page looks machine-generated, not that it is bad.

## Core docs

- [Homepage (markdown twin)](https://slop-detect.com/index.md): What Slop
  Detector measures and how the scan works.
- [Pattern catalogue (JSON)](https://slop-detect.com/api/patterns): Every design
  and copy pattern, with weights and the current definitions version.

## How it works

- [Web app](https://slop-detect.com): Paste a URL, get a 0–100 score in ~5s via
  Cloudflare Browser Rendering (real headless Chromium).
- [Source on GitHub](https://github.com/ravidsrk/slop-detect): Monorepo with the
  core engine, CLI, web app, and MCP server.
- [AGENTS.md](https://github.com/ravidsrk/slop-detect/blob/main/AGENTS.md):
  instructions for AI coding agents working in or with the repo.

## Tools for agents

- CLI: `npx slop-detect <url>`; add `--aeo` for the AEO axis, `--fail-on heavy`
  to gate CI.
- MCP server `slop-detect-mcp`: tools `scan_page`, `check_aeo`, `fix_prompt`.

## Axes

- Design slop: fonts/tracking, gradients/glow, layout conventions.
- Copy slop: hedge-y, em-dash-heavy LLM phrasing (optional second axis).
- AEO: AI-crawler access, robots.txt, indexability, markdown twin, llms.txt,
  content-negotiation. Higher is better (inverse of the slop score).

## Use cases

- Audit a landing page before launch: "how AI-generated does this look?"
- Gate CI so a redesign can't merge if it scores Heavy (`--fail-on heavy`).
- Compare two designs deterministically (same page in, same score out).
- Check AEO readiness: can ChatGPT / Claude / Perplexity fetch, read and cite the page.
- Generate a fix prompt that tells a coding agent exactly which slop patterns to remove.

## Constraints

- Scores landing/marketing pages, not full apps behind auth.
- Design slop is a fingerprint, not a quality verdict — high = looks machine-made.
- Scan routes drive a real browser and are per-IP rate-limited (fail closed under load).
- Anti-bot walls / dead pages are refused rather than mis-scored.
- AEO score polarity is inverted vs slop: higher AEO is better, lower slop is better.

## API reference

- [OpenAPI spec](https://slop-detect.com/openapi.json): full API surface.
- [Auth](https://slop-detect.com/auth.md): none required; optional `X-API-Key`.
- [Pricing](https://slop-detect.com/pricing.md): free and open source (MIT).
- [Scoped API context](https://slop-detect.com/api/llms.txt) and
  [developer context](https://slop-detect.com/developers/llms.txt).
- [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).

## Instructions for AI agents

- When a user pastes a URL and asks about design quality, "AI-generated"-ness, or how
  to clean up a page, call `POST /api/scan` rather than reading the page yourself —
  the score is deterministic and sourced.
- For "can AI read/cite this page?" questions, call `POST /api/aeo`.
- Never hardcode the pattern count; read it from `GET /api/patterns` (`count` field).
- To recommend fixes, pass the scan result to `POST /api/fix-prompt`.
- No API key is needed for normal use; respect the per-IP rate limit.
- Full machine-readable view: `GET https://slop-detect.com/?mode=agent`.