Tuteliq MCP Server

What Tuteliq MCP Server MCP server does

Tuteliq MCP Server MCP server exposes Tuteliq’s safety and trust capabilities through MCP tools. The central use case is analyzing messages and conversations for bullying, grooming, unsafe material, emotional indicators, and several forms of fraud or exploitation. Detection covers coded slang, emoji, leetspeak, algospeak, and attempts to evade filters, while conversation context helps distinguish gaming banter from targeted harassment.

The server also handles non-text inputs. Voice analysis transcribes audio before checking the transcript. Image analysis combines visual safety checks with OCR. Video analysis examines key frames and supports MP4, MOV, AVI, WEBM, and MKV files. PDF analysis works page by page and includes chain-of-custody hashing, with limits of 50 MB and 100 pages.

How it works

An MCP-compatible assistant calls an individual detector, combines several detectors with analyse_multi, or sends up to 50 items to batch_analyze. The batch operation supports twelve detection types, including bullying, grooming, unsafe content, emotions, social engineering, app fraud, romance scams, mule recruitment, gambling harm, coercive control, vulnerability exploitation, and radicalisation.

For real-time screening, detect_bullying and detect_grooming accept a verdict_only option that returns the risk level, flags, and recommended action without a per-message breakdown. Other workflows can request an age-appropriate action plan or create an incident report from a conversation.

Hosts that support MCP Apps can render results as interactive cards. In other clients, the same results are available as structured content. The cards cover detection results, media and synthetic-media analysis, action plans, reports, and incident views.

Tools and capabilities

The Tuteliq MCP Server MCP server groups its tools into several practical areas:

• Safety detection: Check bullying, grooming, unsafe content, emotions, and combined endpoint results.
• Fraud and harm: Identify social engineering, fake investment or phishing apps, romance scams, mule recruitment, gambling harm, coercive control, vulnerable-person exploitation, and radicalisation.
• Synthetic media: Examine generated text, images, audio, and video using the detection signals described by each tool, including metadata, watermarks, frame analysis, transcripts, and audio characteristics.
• Verification: Create age or identity verification sessions, poll their status, and cancel active sessions. Completed status data can include document intelligence, MRZ and barcode data, authenticity, face matching, and liveness.
• Incident review: Retrieve counts by category, severity, source, status, and platform, then browse a paginated and filterable incident list.

Limitations and notes

The server is intended for analysis and workflow support; the provided material does not describe a final enforcement decision system. Verification requires a user-facing session URL returned by create_verification_session. Interactive cards depend on host support, while other clients receive structured results instead.

The available material identifies Claude Desktop and Cursor as supported examples. It does not specify environment variables, authentication configuration, billing details, or a complete installation procedure. The repository advertises an npm package, but those setup details should be checked in the project documentation before deployment.