SafeNest MCP Server

What is this?

Tuteliq MCP Server brings AI-powered child safety tools directly into Claude, Cursor, and other MCP-compatible AI assistants. Ask Claude to check messages for bullying, detect grooming patterns, or generate safety action plans.

Reads context, not just keywords. Every detector understands coded slang, emoji, leetspeak, algospeak, and deliberate filter evasion, and weighs the conversation around a message — so it tells gaming trash-talk apart from targeted harassment instead of drowning your team in false positives. This coded-language resilience is platform-wide (it applies to grooming, fraud, radicalisation, and the rest, not just bullying) and is built from our own research into how bad actors evade moderation. In an internal benchmark of coded-language and filter-evasion cases, Tuteliq detected roughly 1.7x more of them than leading general-purpose moderation APIs (319-case evasion set; vendors unnamed).

Fast mode. Pass verdict_only: true on detect_grooming or detect_bullying to get just the verdict (risk level, flags, recommended action) without the per-message breakdown — lower latency for real-time screening. The verdict itself is unchanged.

Interactive results. In hosts that support MCP Apps, results render as interactive cards rather than walls of JSON — see Interactive widgets below.

Related MCP server: ChuangSi AI Model Content Protection

Interactive widgets

Twelve widgets return a rendered card instead of raw text in hosts that support MCP Apps (Claude desktop and web, and other MCP-compatible clients). Everywhere else the same data arrives as structuredContent, so nothing depends on the UI.

Every card carries the same frame: a chrome bar naming the tool that produced the result, the result itself, and a footer with the data-handling note and a Trust Center link. In a transcript holding a dozen results, the chrome bar is what tells you which is which.

Severity is rankable by colour. The ramp runs monotonically from safe to critical, so two chips can be compared without reading their labels:

Design notes. The widgets are deliberately calm. They report on grooming, self-harm, and abuse, and a card that animates or pulses at the reader turns material that is already distressing into an alarm they cannot dismiss. Severity is carried by a rule and a glyph, not by motion. The crisis-support card leads with reassurance rather than the severity colour, and its helpline numbers are the largest targets on the card because transcribing digits under stress is where people fail.

Widgets are read-only renderers by design. Selecting incidents in the list widget assembles the ID list for a batch_review_incidents call you fire yourself — the mutating call still goes through your host's approval step, so the human-in-the-loop stays in the loop.

Working on the widgets

npm run preview:ui   # builds every widget against fixture data
open dist-preview/__preview.html

Widget source lives in ui/src. Design tokens are centralised in ui/src/theme.ts; prefer them over colour literals so the palette stays in one place.

Available Tools (81 MCP)

Safety Detection

Fraud & Harm Detection

Voice, Image, Video & Document Analysis

Synthetic Content Detection

Identity & Age Verification

Incidents & Moderation

Read the incident store, triage a queue, and record moderator decisions. The review tools emit signed receipts for EU AI Act Art 14 human-oversight evidence.

The decision is the moderator's, and the card makes them take it. The action buttons call review_incident through the host, because a moderator clicking "Escalate" is the human decision. They do not fire on one click: review_incident persists an override and emits a signed Art 12 audit receipt and requires a reason_code, so the button opens a reason picker and a second click commits. Nothing is ever defaulted into that receipt on the moderator's behalf.

The reasoning, confidence and analysis trace on the card are supplied by the calling assistant and are labelled as such — an argument for a human to weigh, not a Tuteliq measurement.

Pass operator_name to brand the header with the customer or team name. Omit it and the card is unbranded — it is never defaulted to a placeholder.

Webhook Management

Pricing

Usage & Billing

Policy Configuration

Policy Automation Rules

Detection Settings

Threat Intelligence (Business+ tier)

GDPR Account

Breach Management

Common Parameters

Context Fields

All detection tools accept an optional context object. These fields influence severity scoring and classification:

sender_trust Behavior

When sender_trust is set to "verified" or "trusted":

• AUTH_IMPERSONATION is fully suppressed — a verified sender cannot be impersonating an authority

• URGENCY_FABRICATION is suppressed for routine time-sensitive information (schedules, deadlines, appointments)

• Content is only flagged if it contains genuinely malicious elements (credential theft, phishing links, financial demands)

• This prevents false positives on legitimate institutional messages (school notifications, hospital reminders, government advisories)

support_threshold

Controls when crisis support resources (helplines, text lines, web resources) are included in the response:

Note: Critical severity always includes support resources regardless of the threshold setting.

analyse_multi Endpoint Values

The analyse_multi tool accepts up to 10 endpoints per call. Valid endpoint values:

Installation

Tuteliq is a hosted MCP server at https://api.tuteliq.ai/mcp. Most clients should connect with OAuth and install nothing.

Connect with OAuth (recommended)

Point the client at the URL with no credentials and sign in through the browser. Tuteliq implements OAuth 2.1 with dynamic client registration and PKCE, so the client registers itself. Nothing is pasted into a config file, and access is revoked from the dashboard rather than by editing your machine.

Claude Desktop: Settings > Connectors, Add custom connector, name it Tuteliq, URL https://api.tuteliq.ai/mcp, then Connect and approve in the browser.

Claude Code, Cursor, Windsurf and other clients supporting remote servers:

{
  "mcpServers": {
    "tuteliq": {
      "type": "http",
      "url": "https://api.tuteliq.ai/mcp"
    }
  }
}

In Claude Code, run /mcp to start the sign-in if it does not open on its own.

Static token (headless and automation)

OAuth needs a browser, so a CI pipeline, cron job or container cannot complete it. Send a token in the Authorization header instead, generated in the dashboard under Settings > Plugins. This is a long-lived credential: keep it out of version control, and prefer OAuth wherever a browser exists.

{
  "mcpServers": {
    "tuteliq": {
      "type": "http",
      "url": "https://api.tuteliq.ai/mcp",
      "headers": {
        "Authorization": "Bearer your-secure-token"
      }
    }
  }
}

stdio (clients without remote server support)

For clients that only speak stdio. This runs a local process that calls the same hosted API, so the tools are identical; only the transport and authentication differ.

{
  "mcpServers": {
    "tuteliq": {
      "command": "npx",
      "args": ["-y", "@tuteliq/mcp"],
      "env": {
        "TUTELIQ_API_KEY": "your-api-key"
      }
    }
  }
}

Usage Examples

Once configured, you can ask Claude:

Bullying Detection

"Check if this message is bullying: 'Nobody likes you, just go away'"

Response:

## ⚠️ Bullying Detected

**Severity:** 🟠 Medium
**Confidence:** 92%
**Risk Score:** 75%

**Types:** exclusion, verbal_abuse

### Rationale
The message contains direct exclusionary language...

### Recommended Action
`flag_for_moderator`

Grooming Detection

"Analyze this conversation for grooming patterns..."

Quick Safety Check

"Is this message safe? 'I don't want to be here anymore'"

Emotion Analysis

"Analyze the emotions in: 'I'm so stressed about school and nobody understands'"

Action Plan

"Give me an action plan for a 12-year-old being cyberbullied"

Incident Report

"Generate an incident report from these messages..."

Voice Analysis

"Analyze this audio file for safety: /path/to/recording.mp3"

Image Analysis

"Check this screenshot for harmful content: /path/to/screenshot.png"

Webhook Management

"List my webhooks" "Create a webhook for critical incidents at https://example.com/webhook"

Usage

"Show my monthly usage"

Synthetic Content Detection

"Is this image AI-generated? /path/to/suspect-image.jpg" "Check if this audio is a voice clone: /path/to/voice.mp3" "Analyze this video for deepfake indicators: /path/to/video.mp4" "Is this text AI-generated? 'The generated text to analyze...'" "Show me the synthetic content profile for customer cust_xyz789"

Identity & Age Verification

"Create an age verification session" "Create an identity verification session with passport as preferred document" "Check the status of verification session abc123" "Cancel verification session abc123"

Fraud Detection

"Check this message for social engineering: 'Your account will be suspended unless you verify now'" "Is this a romance scam? 'I know we just met online but I need help with a medical bill'"

Get Started (Free)

1. Create a free Tuteliq account

2. Go to your Dashboard and generate an API Key

3. For Claude Desktop and other MCP plugins, generate a Secure Token under Settings > Plugins

4. Use the API key for direct API/SDK access, or the Secure Token when connecting via MCP

Requirements

• Node.js 18+

• Tuteliq API key

Supported Languages (27)

Language is auto-detected when not specified. Beta languages have good accuracy but may have edge cases compared to English.

Best Practices

Message Batching

The bullying and unsafe content tools analyze a single text field per request. If you're analyzing a conversation, concatenate a sliding window of recent messages into one string rather than sending each message individually. Single words or short fragments lack context for accurate detection and can be exploited to bypass safety filters.

The grooming tool already accepts a messages[] array and analyzes the full conversation in context.

PII Redaction

Enable PII_REDACTION_ENABLED=true on your Tuteliq API to automatically strip emails, phone numbers, URLs, social handles, IPs, and other PII from detection summaries and webhook payloads. The original text is still analyzed in full — only stored outputs are scrubbed.

Supported Languages

Tuteliq supports 27 languages with automatic detection — no configuration required.

English (stable) and 26 beta languages: Spanish, Portuguese, Ukrainian, Swedish, Norwegian, Danish, Finnish, German, French, Dutch, Polish, Italian, Turkish, Romanian, Greek, Czech, Hungarian, Bulgarian, Croatian, Slovak, Lithuanian, Latvian, Estonian, Slovenian, Maltese, and Irish.

All 24 EU official languages + Ukrainian, Norwegian, and Turkish. Each language includes culture-specific safety guidelines covering local slang, grooming patterns, self-harm coded vocabulary, and filter evasion techniques.

See the Language Support docs for details.

Support

• API Docs: docs.tuteliq.ai

• Discord: discord.gg/7kbTeRYRXD

• Email: support@tuteliq.ai

Privacy & Legal

Tuteliq processes content for safety analysis on behalf of the operator (the API key holder). The MCP server is a thin transport that forwards requests to api.tuteliq.ai over TLS — no text, audio, image, or video content is stored locally by the MCP package.

What is collected, used, and stored

• Authentication: API keys (server-side) or OAuth 2.1 access tokens (Claude / Cursor connectors). OAuth tokens are issued by api.tuteliq.ai and follow the standard RFC 9728 / RFC 8414 discovery flow.

• Request content: text, audio, images, video, and PDFs you submit to detection or analysis tools are processed in-memory by the upstream API. Content is not retained beyond the request unless you explicitly enable history features in the dashboard.

• Metadata stored: request timestamps, tool name, status, latency, and credit consumption — used for usage analytics, billing, and audit logs.

• PII redaction: enable PII_REDACTION_ENABLED=true to strip emails, phone numbers, URLs, social handles, and IPs from stored summaries and webhook payloads. The original input is still analyzed in full; only stored outputs are scrubbed.

• Sub-processors and retention: see the DPA.

• Your rights: the MCP exposes GDPR tools (export_account_data, delete_account_data, record_consent, withdraw_consent, rectify_data, get_audit_logs) so you can exercise data subject rights directly from your client.

License

MIT License - see LICENSE for details.

Get Certified — Free

Tuteliq offers a free certification program for anyone who wants to deepen their understanding of online child safety. Complete a track, pass the quiz, and earn your official Tuteliq certificate — verified and shareable.

Three tracks available:

Start here → tuteliq.ai/certify

• 100% Free — no login required

• Verifiable certificate on completion

• Covers grooming recognition, sextortion, cyberbullying, regulatory obligations (KOSA, EU DSA), and more

The Mission: Why This Matters

Before you decide to contribute or sponsor, read these numbers. They are not projections. They are not estimates from a pitch deck. They are verified statistics from the University of Edinburgh, UNICEF, NCMEC, and Interpol.

• 302 million children are victims of online sexual exploitation and abuse every year. That is 10 children every second. (Childlight / University of Edinburgh, 2024)

• 1 in 8 children globally have been victims of non-consensual sexual imagery in the past year. (Childlight, 2024)

• 370 million girls and women alive today experienced rape or sexual assault in childhood. An estimated 240–310 million boys and men experienced the same. (UNICEF, 2024)

• 29.2 million incidents of suspected child sexual exploitation were reported to NCMEC's CyberTipline in 2024 alone — containing 62.9 million files (images, videos). (NCMEC, 2025)

• 546,000 reports of online enticement (adults grooming children) in 2024 — a 192% increase from the year before. (NCMEC, 2025)

• 1,325% increase in AI-generated child sexual abuse material reports between 2023 and 2024. The technology that should protect children is being weaponized against them. (NCMEC, 2025)

• 100 sextortion reports per day to NCMEC. Since 2021, at least 36 teenage boys have taken their own lives because they were victimized by sextortion. (NCMEC, 2025)

• 84% of reports resolve outside the United States. This is not an American problem. This is a global emergency. (NCMEC, 2025)

End-to-end encryption is making platforms blind. In 2024, platforms reported 7 million fewer incidents than the year before — not because abuse stopped, but because they can no longer see it. The tools that catch known images are failing. The systems that rely on human moderators are overwhelmed. The technology to detect behavior — grooming patterns, escalation, manipulation — in real-time text conversations exists right now. It is running at api.tuteliq.ai.

The question is not whether this technology is possible. The question is whether we build the company to put it everywhere it needs to be.

Every second we wait, another child is harmed.

We have the technology. We need the support.

If this mission matters to you, consider sponsoring our open-source work so we can keep building the tools that protect children — and keep them free and accessible for everyone.