vielhuber/slopdetector

Provider-based PHP helper and MCP server for AI-text and plagiarism detection.

1.0.2 2026-08-23 05:13 UTC

Requires

• php: ^8.5
• ext-curl: *
• ext-json: *
• vielhuber/simplemcp: ^1.1.4

Requires (Dev)

• phpunit/phpunit: ^13

MIT

Published on Packagist.org on 2026-08-23 05:14 UTC

README

🍲slopdetector🍲

slopdetector is a provider-based PHP helper and MCP server for AI-text and plagiarism detection.

the first provider is Pangram. additional providers can implement the small Provider interface without changing the public PHP or MCP API.

slopdetector requires PHP 8.5 or newer.

installation

composer require vielhuber/slopdetector

configuration

copy .env.example to .env or provide the same variables through the process environment:

PROVIDER=pangram
PANGRAM_API_KEY=your_api_key
MCP_TOKEN=

the API key is sent only in Pangram's x-api-key request header. text and files are transferred to Pangram for analysis. public dashboard links are disabled unless explicitly requested.

PHP

<?php
declare(strict_types=1);

require __DIR__ . '/vendor/autoload.php';

use vielhuber\slopdetector\slopdetector;

$detector = slopdetector::create();
$models = $detector->listModels();

$result = $detector->analyzeText(
    text: 'Text to analyze',
    model: $models[0]
);

print_r($result);

example response:

{
    "stage": "STAGE_SUCCESS",
    "text": "Text to analyze",
    "version": "3.3.2",
    "headline": "Human Written",
    "prediction": "We believe this is human-written",
    "prediction_short": "Human",
    "fraction_ai": 0,
    "fraction_ai_assisted": 0,
    "fraction_human": 1,
    "num_ai_segments": 0,
    "num_ai_assisted_segments": 0,
    "num_human_segments": 1,
    "windows": [
        {
            "text": "Text to analyze",
            "label": "Human Written",
            "ai_assistance_score": 0.01,
            "confidence": "High",
            "start_index": 0,
            "end_index": 15,
            "word_count": 3,
            "token_length": 3
        }
    ]
}

models are discovered dynamically. analyzeText() and submitBulk() deliberately require an explicit model so the integration remains valid when Pangram makes model selection mandatory after September 30, 2026. Pangram's file endpoint currently uses its server-side default model and therefore has no model argument.

supported file formats are PDF, DOCX, and RTF:

$result = $detector->analyzeFile(filePath: '/path/to/document.pdf');

bulk jobs remain asynchronous:

$job = $detector->submitBulk(
    items: [
        ['id' => 'article-1', 'text' => 'First text'],
        ['id' => 'article-2', 'text' => 'Second text']
    ],
    model: $models[0]
);

$status = $detector->getBulkStatus(bulkId: $job['bulk_id']);
$results = $detector->getBulkResults(bulkId: $job['bulk_id'], offset: 0, limit: 100);

providers

providers implement vielhuber\slopdetector\Provider. the selected provider is configured through PROVIDER; currently supported: pangram.

MCP server

vendor/bin/mcp-server.php

available tools:

• list_models
• analyze_text
• check_plagiarism
• analyze_file
• submit_bulk
• get_bulk_status
• get_bulk_results

analyze_text hides Pangram's internal submit/poll cycle and returns the completed result. bulk processing exposes separate submit, status, and paginated result tools because jobs can be long-running.

tests

composer install
vendor/bin/phpunit

the test suite uses a fake transport and does not send texts, files, or API credentials to Pangram.