What's New

Fraud & Harm Detection (8 new tools)

• detect_social_engineering — Pretexting, urgency fabrication, authority impersonation
• detect_app_fraud — Fake investment platforms, phishing apps, subscription traps
• detect_romance_scam — Love-bombing, financial requests, identity deception
• detect_mule_recruitment — Easy-money offers, bank account sharing tactics
• detect_gambling_harm — Chasing losses, concealment, gambling-related distress
• detect_coercive_control — Isolation, financial control, monitoring, threats
• detect_vulnerability_exploitation — Targeting elderly, disabled, or financially distressed individuals
• detect_radicalisation — Extremist rhetoric, us-vs-them framing, ideological grooming

Crisis Support Card

• Country-specific crisis helplines displayed in the MCP widget when high/critical threats are detected
• Warm, category-aware messages (self-harm, bullying, grooming, etc.)
• Phone numbers copy to clipboard on click with confirmation feedback
• Resource links open in system browser via MCP Apps app.openLink() API

Language Support

• 27 languages now supported with auto-detection
• English (stable) + 26 beta languages including Spanish, Portuguese, French, German, Italian, Swedish, Norwegian, Finnish, Ukrainian, and more

UI Improvements

• New components: SupportCard, BrandedLoader, FooterLinks, UpsellBanner
• All widget links now work correctly in Claude Desktop via app.openLink()

Security

• Fixed Hono prototype pollution vulnerability (GHSA-v8w9-8mx6-g223)
• Removed internal infrastructure references from tool descriptions

Other

• Updated README with all 41 MCP tools, fraud section, languages table
• Verification tools documented as API & SDK only
• Added .claude/ to .gitignore

Full Changelog: v3.2.0...v3.3.4