TruthScan

@TruthScan analyze this press release draft for AI-generated text and show me which sections were likely written by a model

Analysis complete. The 847-word draft shows 73% AI-generated probability overall. Three sections flagged high confidence (>85%): the executive quote in paragraph 2, the product benefits list in paragraph 4, and the boilerplate closing. The opening anecdote and market-context paragraphs appear human-written (22% and 31% probability). Detection consumed 1,000 credits; your account has 24,000 remaining.

This example demonstrates text analysis with probability scoring across document sections. Each detection consumes credits from your account balance—check remaining quota with the credits tool before batch operations. TruthScan returns probabilities, not certainties; use results as one signal in editorial review, not as definitive proof of authorship.

@TruthScan check if this product photo from our supplier has been AI-generated or manipulated, and download the heatmap overlay

Image analysis complete (Job ID: img_9k2mX). Detection confidence: 91% AI-generated. Primary indicators: unnatural lighting gradients in the background, inconsistent shadow angles, and texture artifacts around product edges. Heatmap downloaded to your project files—red zones highlight the manipulated regions (background 94% confidence, left edge 87%). Analysis cost 500 credits.

This workflow pairs image detection with visual evidence via heatmap download. The MCP submits the job, polls until complete, then fetches the annotated overlay as a managed file. Useful for vendor audits or content moderation pipelines. Note that detection is asynchronous—initial submission returns a job ID you can query later if polling times out.

@TruthScan analyze the last three video testimonials in our content folder and summarize which ones show signs of deepfake or face-swapping

Analyzed three videos (total 4m 32s). Video 1 (customer_testimonial_A.mp4): 8% deepfake probability, no face-swap detected—appears authentic. Video 2 (testimonial_B.mp4): 67% face-swap confidence around the mouth region from 0:12–0:34, likely synthetic. Video 3 (testimonial_C.mp4): 94% AI-generated probability, multiple facial inconsistencies detected. Videos 2 and 3 warrant manual review before publication. Total cost: 6,000 credits (2,000 per video).

This synthesis example shows the MCP analyzing multiple assets and the AI summarizing risk across a batch. Video detection is credit-intensive and non-idempotent—resubmitting the same URL consumes credits again. Best practice: store job IDs and use the get-result tool to retrieve completed analyses without re-charging your account.

When TruthScan fits a 3-person investigative desk

A small newsroom gets 20-30 reader-submitted tips per week—photos, videos, audio clips claiming misconduct or breaking news. TruthScan is the right call here because the team needs a single API to triage synthetic media across formats before investing reporter hours. The 8-tool spread (image, video, audio, text, PDF) covers the inbound mix, and the credit-based pricing maps cleanly to per-tip budgeting. The threshold: if your volume climbs past 200 submissions per week, the bounded-polling design starts to feel slow and you'll want batch endpoints TruthScan doesn't expose. Below that line, the MCP saves the manual export-upload-check loop and keeps the triage workflow inside one shared workspace.

TruthScan works for small-case tampering checks

A 2-lawyer practice handling employment disputes needs to verify that opposing counsel's submitted PDFs and image exhibits haven't been altered. TruthScan's PDF and image analyzers catch obvious tampering without sending files to a third-party forensics vendor. The 2 MB PDF limit and per-page credit cost (1,000 credits per page) make this viable for cases with under 100 pages of contested exhibits. Beyond that threshold, the credit burn gets expensive and the lack of batch processing becomes a bottleneck. For small-case work—10 to 50 documents per matter—the MCP keeps the verification step in-house and gives both attorneys audit trails via the shared Switchy workspace.

When TruthScan handles low-volume community posts

A 5-person volunteer team moderating a regional forum sees 30-50 user-submitted posts per day, mixing text, images, and occasional video. TruthScan is borderline here: the multi-format coverage is perfect, but the credit model and bounded polling mean the team can't auto-scan every post in real time. The right pattern is selective scanning—flag posts that trip keyword or report thresholds, then run TruthScan on those 5-10 daily outliers. If your moderation queue exceeds 100 items per day or you need sub-second response, TruthScan's async design won't keep pace. Below that line, the MCP gives moderators a shared detection layer without building custom integrations for each media type.