Phish Radar - Phishing Detection & Brand Monitoring MCP Server

Real-time phishing detection and brand monitoring system for AI agents. Analyzes URLs and domains for phishing indicators, typosquatting, homoglyphs, and lookalike domains.

Features

1. Phishing Detection (phishing_detect)

Comprehensive URL/domain analysis with:

• Typosquatting detection - Identifies domains similar to known brands (Levenshtein distance)
• Homoglyph detection - Catches visually similar character substitutions (0/O, 1/l/I, etc.)
• Domain age analysis - Detects recently registered suspicious domains
• SSL certificate inspection - Flags free certificates (Let's Encrypt) when combined with brand similarity
• DNS records - Checks for MX, SPF, DKIM, DMARC records
• Hosting location - Identifies suspicious geography patterns

Score: 0-100 phishing confidence

• PHISHING (70+): Block immediately
• SUSPICIOUS (40-69): Review recommended
• SAFE (<40): Low risk

2. Domain Trust Score (domain_trust)

Evaluates domain trustworthiness (0-100) based on:

• Email infrastructure (MX, SPF, DKIM, DMARC) - 25 points
• SSL certificate type (EV > OV > DV) - 20 points
• TLD reputation - 20 points
• Nameserver reputation - 15 points
• DNS records - 5 points

Levels:

• HIGHLY_TRUSTED (80+): Enterprise-grade domain
• TRUSTED (60-79): Legitimate domain
• NEUTRAL (40-59): Unproven domain
• SUSPICIOUS (20-39): Warning signs present
• UNTRUSTED (<20): High-risk domain

3. Brand Monitor (brand_monitor)

Detects lookalike domains targeting your brand:

• Typosquats - Single character swaps/deletions (paypa1.com vs paypal.com)
• Homoglyphs - Visually confusing substitutions
• Keyword additions - Security-related additions (paypal-secure.com)
• TLD variations - Same domain on different extensions (paypal.xyz)

Returns high-risk domains that actually exist with full risk assessment.

Usage

Give the Actor a list of domains or URLs in targets and pick a mode. Up to 50 targets per run; anything beyond that is reported in a notice row instead of being dropped silently.

Targets may be bare domains (stripe.com) or full URLs (https://www.stripe.com/login) — URLs are reduced to the hostname before lookup. If your client sends the list under a different key, it is still read: target, url, urls, startUrls, domain, domains, brand_domain, query. The legacy MCP shape {"toolName": ..., "input": {...}} keeps working unchanged.

Example 1: Full report (default)

full runs phishing detection and trust scoring on the same target and merges them into one row. The console example run ("Example run input") prefills 12 targets so a fresh start returns a useful multi-row report rather than a single sample; pass your own list in targets to override it.

{
  "targets": ["paypa1-secure.com", "stripe.com", "github.com"],
  "mode": "full"
}

One row per target:

{
  "tool": "full",
  "url": "paypa1-secure.com",
  "domain": "paypa1-secure.com",
  "verdict": "SUSPICIOUS",
  "confidence": 0.95,
  "phishing_score": 60,
  "signals": {
    "typosquat_of": "paypal.com",
    "typosquat_distance": 1,
    "homoglyph_detected": false,
    "homoglyph_count": 0,
    "domain_age_days": null,
    "ssl_issuer": null,
    "ssl_free_cert": false,
    "has_mx_records": false,
    "suspicious_tld": false,
    "dns_a_records": [],
    "hosting_country": "UNKNOWN",
    "domain_exists": false,
    "has_ssl": false
  },
  "action": "REVIEW",
  "details": "Suspicious domain characteristics detected (score: 60). Resembles legitimate domain paypal.com.",
  "trust_score": 20,
  "trust_level": "SUSPICIOUS",
  "trust_factors": {
    "has_mx": false,
    "has_spf": false,
    "has_dkim": false,
    "has_dmarc": false,
    "ssl_type": "NONE",
    "ssl_issuer": "None",
    "has_ssl": false,
    "tld_risk": "LOW",
    "nameservers": [],
    "a_records_count": 0,
    "resolves": false
  },
  "timestamp": "2026-09-10T07:46:54.982Z"
}

domain_age_days is always null — the Actor does not query WHOIS. hosting_country is a coarse IP-range guess, not a GeoIP lookup.

Example 2: Domain trust only

{
  "targets": ["github.com"],
  "mode": "domain_trust"
}

Response row:

{
  "tool": "domain_trust",
  "domain": "github.com",
  "trust_score": 100,
  "trust_level": "HIGHLY_TRUSTED",
  "factors": {
    "has_mx": true,
    "has_spf": true,
    "has_dkim": true,
    "has_dmarc": true,
    "ssl_type": "DV",
    "ssl_issuer": "Sectigo Public Server Authentication CA DV E36",
    "has_ssl": true,
    "tld_risk": "LOW",
    "nameservers": ["ns-1283.awsdns-32.org", "dns1.p08.nsone.net"],
    "a_records_count": 4,
    "resolves": true
  },
  "timestamp": "2026-09-10T07:48:11.204Z"
}

Example 3: Monitor brand lookalikes

brand_monitor returns one row per live lookalike domain, sorted with the highest risk first.

{
  "targets": ["stripe.com"],
  "mode": "brand_monitor",
  "lookalikeLimit": 20
}

Response rows:

{
  "tool": "brand_monitor",
  "brand": "stripe.com",
  "brand_name": "stripe",
  "lookalikes_found": 8,
  "domain": "5tripe.com",
  "risk": "HIGH",
  "squat_type": "homoglyph",
  "tld_risk": "LOW",
  "exists": true,
  "timestamp": "2026-09-10T07:52:03.771Z"
}

Only domains that currently resolve in DNS are returned — parked and unregistered candidates are dropped.

Legacy MCP format

The original nested format still works unchanged:

{
  "toolName": "domain_trust",
  "input": { "domain": "example.com" }
}

Implementation Details

No External API Dependencies

• DNS Queries: Native Node.js dns/promises module
• SSL Certificates: Node.js tls module for direct connections
• WHOIS: Direct TCP port 43 protocol (future enhancement)
• GeoIP: Simplified IP-to-country mapping (for production, integrate MaxMind)

Known Brands Database

Includes 80+ major brands across:

• Financial Services (PayPal, Stripe, Square, etc.)
• Tech Companies (Apple, Google, Microsoft, etc.)
• Cryptocurrency (Coinbase, Kraken, Binance, etc.)
• Cloud Providers (AWS, Azure, Google Cloud, etc.)
• E-commerce (Amazon, eBay, Shopify, etc.)

Easily extensible in src/data/known_brands.js.

Suspicious TLDs

Monitors 40+ high-risk TLDs commonly used in phishing:

• Cheap registration TLDs: .xyz, .top, .click, .bid
• Free TLDs: .tk, .ml, .ga, .cf
• Cryptocurrency scams: .btc, .eth, .crypto
• Sanctioned regions: .ru, .ir, .kp

Phishing Score Calculation

High-Risk Signals (Add Points)

• Typosquat distance 1: +35 points
• Typosquat distance 2: +25 points
• Homoglyph detected: +20 points
• Suspicious TLD: +15 points
• Free SSL + typosquat: +15 points
• No MX + typosquat: +10 points
• Hosting in RU + typosquat: +10 points

Low-Risk Signals (Subtract Points)

• Has MX records: -5 points
• Trusted CA cert: -5 points

Final Score Clamped: 0-100

Confidence Levels

• 70+ (PHISHING): 0.70-0.99 confidence → BLOCK
• 40-69 (SUSPICIOUS): 0.50-0.95 confidence → REVIEW
• <40 (SAFE): 0.95+ confidence → ALLOW

Pricing

• $0.05 per target analyzed — one tool-call event per entry in targets.
• A run with 10 domains costs $0.50, whatever mode you choose.
• brand_monitor charges per brand, not per lookalike: one brand returning 20 lookalike rows is still $0.05.
• Status rows (for example "no target supplied") are never charged.
• Each target gets 25 seconds. A target that does not answer DNS or TLS in time returns an ERROR row saying so, and that row is charged like any other delivered result — it keeps one dead target from starving the rest of your list.

Implementation Notes

Security Considerations

• No WHOIS PII collection: Personal information (owner names, emails) excluded per GDPR
• No certificate validation bypass: Uses system trust stores
• Rate limiting: Implement externally to avoid DNS query floods
• Timeout protection: 3-second SSL connection timeout, 3-second DNS query timeout, 25-second per-target deadline

Performance

• Parallel DNS queries for speed
• Batch processing for large domain lists (10 concurrent)
• Caching at application level (not in Actor)

Future Enhancements

1. WHOIS Integration: For domain registration date analysis
2. GeoIP Database: MaxMind for accurate IP location
3. SSL Certificate History: Certificate Transparency logs
4. Reputation Scoring: Integration with Phishing/Malware databases
5. Machine Learning: Pattern recognition for new phishing techniques

API Signature

Request format

{
  "targets": ["example.com"],
  "mode": "full|phishing_detect|domain_trust|brand_monitor",
  "lookalikeLimit": 20
}

The MCP format (toolName + nested input) is still accepted.

Error row

A target that fails analysis still returns a row, so you always know what happened to each entry:

{
  "tool": "full",
  "domain": "example.com",
  "verdict": "ERROR",
  "error": "Error description",
  "timestamp": "2026-09-10T07:48:11.204Z"
}

Testing

npm install

# Local run with a fixed input
mkdir -p storage/key_value_stores/default
echo '{"targets":["paypa1-secure.com"],"mode":"full"}' > storage/key_value_stores/default/INPUT.json
CRAWLEE_STORAGE_DIR=./storage node src/main.js
cat storage/datasets/default/*.json

# On the platform
apify call phish-radar --input input.json

Maintenance

• Update known_brands.js quarterly with emerging platforms
• Monitor suspicious_tlds.js for new high-abuse extensions
• Review phishing signal weights annually based on threat patterns

Support

For brand additions or TLD updates, modify:

• Brand list: src/data/known_brands.js
• TLD list: src/data/suspicious_tlds.js

---

Version: 1.0 Node.js: 18+ License: Private (ntriq)

---

🔗 Related Actors by ntriqpro

Build your data pipeline with the ntriqpro Actor suite:

• maigret-actor — Maigret OSINT — Username search
• website-tech-detector — Website Tech Stack Scanner
• eu-sanctions-monitor — EU Sanctions Screening — KYC/AML

⭐ Love it? Leave a Review

Your rating helps professionals discover this actor. Rate it here.

Disclaimer

RISK ASSESSMENT DISCLAIMER: This tool provides data-driven risk analysis based on publicly available and curated sources. Assessments are indicative only and NOT professional risk management advice. You must engage qualified professionals (risk managers, insurance brokers, compliance officers) to validate findings before making investment or operational decisions. Risk scores do not constitute investment advice or due diligence completion. We assume no liability for business decisions based on this tool's output.