# Tuputech Moderation MCP > Tuputech Moderation provides your agent 10 specialized endpoints to check content safety across text, images, audio, and video streams. It detects deepfakes, spam, abuse, NSFW material, advertisements, and more automatically via AI scanning. Use it when you need guaranteed policy enforcement on user-submitted media. ## Overview - **Category:** image-video - **Price:** Free - **Tags:** content-moderation, deepfake-detection, anti-spam, visual-recognition, trust-and-safety ## Description **Tuputech Moderation gives your agent ten specialized endpoints to check content safety across text, images, audio, and video streams.** You use this when you need hard policy enforcement on anything users submit. It handles deepfakes, spam, abuse, NSFW material, ads, and more automatically through AI scanning. For text analysis, you can run checks using `scan_text_abuse` to flag blocks of language that contain abusive, toxic, or hate speech patterns. If your policy is strict about political content, `scan_text_politics` checks submitted text against specific guidelines. You'll use `scan_text_spam` when you need to spot commercial advertisements or common spam patterns within a written block. When dealing with images, the server offers several deep checks. To determine if an image is synthetic—meaning it was generated by AI—you call `scan_image_synthetic`. You can scan for explicit material using `scan_image_porn`, which flags Not Safe For Work (NSFW) content. If you're looking for severe violations, `scan_image_terrorism` scans images specifically for extreme gore or terrorist-related material. To filter out disguised ads, use `scan_image_ad`; this tool detects advertisements embedded within an image file. For large media files, the system handles streams asynchronously so your workflow doesn't time out. You trigger a background scan on video content using `scan_video_stream`. For audio files or streams, you start the check with `scan_audio_stream`. These stream functions are critical for processing big batches of media. Finally, if the AI scanning results aren't perfect, you can help refine its accuracy by submitting manual feedback via `submit_feedback`. ## Tools ### scan_audio_stream Starts an asynchronous scan on a provided audio file or stream. ### scan_image_ad Detects advertisements that are embedded or disguised within an image. ### scan_image_porn Checks a single image for Not Safe For Work (NSFW) or pornographic content. ### scan_image_synthetic Determines if an image was created by AI, flagging potential deepfakes. ### scan_image_terrorism Scans images for content related to terrorism or extreme gore. ### scan_text_abuse Flags text blocks containing abusive, toxic, or hate speech language. ### scan_text_politics Checks if submitted text violates strict political guidelines or policy rules. ### scan_text_spam Scans a block of text to identify spam patterns and commercial advertisements. ### scan_video_stream Starts an asynchronous scan on a video file or stream for policy violations. ### submit_feedback Allows manual feedback submission to help refine the accuracy of AI scanning results. ## Prompt Examples **Prompt:** ``` Check if this profile photo contains NSFW or Gore imagery: https://... ``` **Response:** ``` I issued queries to both `scan_image_porn` and `scan_image_terrorism`. The photo passed all safety thresholds successfully with a 99% confidence score. ``` **Prompt:** ``` Scan this text and see if it is spam or political botting: 'Buy followers here...' ``` **Response:** ``` I dispatched the string to Tuputech. It returned a 'Reject' flag, labeling the content strictly under 'Spam/Fraud'. ``` ## Capabilities ### Scan for Synthetic Media Checks images against deepfake models to determine if they were generated by artificial intelligence. ### Identify Explicit or Violent Content Scans images for NSFW (Not Safe For Work) content, gore, or materials related to terrorism. ### Moderate Text Abuse and Toxicity Analyzes blocks of text to flag abusive language, hate speech, or toxic patterns. ### Filter Spam and Advertisements Scans both text blocks and images to spot spam campaigns or embedded ads. ### Analyze Live Media Streams (Audio/Video) Triggers background scans on video or audio files, which is useful for large content moderation tasks. ## Use Cases ### Reviewing User Profile Photos A user uploads a new profile avatar. The agent runs `scan_image_porn` and `scan_image_terrorism`. If either tool returns a high-risk flag, the system rejects the upload immediately, preventing inappropriate content from ever being visible. ### Filtering Forum Comment Threads A new batch of comments arrives. The agent first runs `scan_text_spam` to catch bots and then uses `scan_text_abuse` on the remaining text. This filters out both commercial junk and hostile language in one pass. ### Monitoring Live Video Broadcasts The platform needs real-time safety checks on streamed video content. The agent initiates `scan_video_stream`, which processes the large feed asynchronously, providing a flag if it detects policy violations like gore or terrorism. ### Processing Multi-Format Submissions A user submits an article that includes text, a photo, and a voice note. The agent executes `scan_text_politics` on the body copy, runs `scan_image_synthetic` on the photo, and uses `scan_audio_stream` on the voice note to ensure all three components pass safety checks. ## Benefits - Deepfake detection is instant. Use `scan_image_synthetic` to check any photo URL and confirm if it's AI-generated before letting it go live. - It handles high volume and complex media types. You can run asynchronous moderation on large videos or audio streams using `scan_video_stream` or `scan_audio_stream` without timing out your workflow. - You get granular risk scoring. Instead of one vague 'flag', the agent runs specific tools like `scan_text_abuse` and `scan_image_porn` to tell you exactly what policy was violated. - It catches hidden content. Use `scan_image_ad` to detect advertisements that are disguised within otherwise innocuous images, which standard filters miss. - The tool covers multiple dimensions at once. You can combine a text check (`scan_text_spam`) with an image check (`scan_image_porn`) in one agent call for comprehensive safety review. ## How It Works The bottom line is you get a reliable, multi-modal safety layer that checks content across text, images, and streams without requiring local compute power. 1. Install the Tuputech Moderation MCP Server. 2. Obtain your Access Key and generate an RSA Private Key from the developer console. 3. Input these credentials into the server. Vurb handles the complex hashing process so your agent just needs to call the tool. ## Frequently Asked Questions **How do I check if an image is fake using scan_image_synthetic?** You pass the image URL to `scan_image_synthetic`. The tool returns a verdict that tells you if the picture was generated by AI. This is crucial for catching deepfakes. **What's the difference between scan_text_abuse and scan_text_spam?** `scan_text_abuse` targets hate speech, harassment, and toxic language. `scan_text_spam` focuses on identifying commercial junk or fraudulent link patterns. **Can I check a video file with the Tuputech Moderation MCP Server?** Yes. You use `scan_video_stream`. This tool initiates an asynchronous scan, which is necessary because videos are large files that require background processing. **Do I need to run all ten tools for moderation?** No. Your agent should select the right combination of tools based on context. For example, if it's a forum post, you only need `scan_text_abuse` and `scan_text_spam`. **What credentials do I need to set up the Tuputech Moderation MCP Server?** You must provide an Access Key (SecretId) and generate your RSA Private Key from the Tuputech Developer Console. Vurb handles the complex underlying RSA hashing, so you don't have to worry about crypto logic in your agent. **Does running `scan_audio_stream` block my AI client while it processes?** No, `scan_audio_stream` runs asynchronously. This means the scan starts and reports back when done, letting your agent continue processing other tasks immediately. **How does the `scan_image_ad` tool detect disguised advertisements?** The tool specifically targets images that contain embedded or subtle advertising content. It flags these deceptive placements so you can filter out hidden spam ads automatically. **If a scan result seems inaccurate, how do I help improve the system using `submit_feedback`?** You use `submit_feedback` to manually correct or flag questionable results. This data directly feeds back into Tuputech's systems, helping refine and improve future AI moderation models. **How is the RSA authentication performed?** No manual prompt instructions are needed. You simply set your RSA Private Key safely into your MCP profile in Vurb. When your AI triggers an endpoint like `scan_image_porn`, the Vurb NodeJS shell dynamically computes the RSA `signature` hashing your exact Nonce+Timestamp completely isolated from your prompt traces.