URL Filter

Advanced URL detection and filtering guardrail that prevents access to unauthorized domains. Uses comprehensive regex patterns and robust URL parsing to detect various URL formats, validates them against security policies, and filters based on a configurable allow list.

Key Security Features:

• Prevents credential injection attacks (user:pass@domain)
• Blocks typosquatting and look-alike domains
• Restricts dangerous schemes (javascript:, data:)
• Supports IP addresses and CIDR ranges
• Configurable subdomain matching

Supported stages

Stage	Supported	Notes
pre_flight	✅	Block disallowed URLs in user messages before the LLM call
input	✅	Runs concurrently with the LLM call
output	✅	Recommended. Prevents the LLM from generating or echoing disallowed URLs in its response

Configuration

{
    "name": "URL Filter",
    "config": {
        "url_allow_list": ["example.com", "192.168.1.100", "https://api.service.com/v1"],
        "allowed_schemes": ["https"],
        "block_userinfo": true,
        "allow_subdomains": false
    }
}

Parameters

• url_allow_list (optional): List of allowed domains, IP addresses, CIDR ranges, or full URLs.

  • Default: [] (blocks all URLs)

• allowed_schemes (optional): Set of allowed URL schemes/protocols.

  • Default: ["https"] (HTTPS-only for security)

• block_userinfo (optional): Whether to block URLs containing userinfo (user:pass@domain) to prevent credential injection attacks.

  • true (default): Blocks URLs containing userinfo
  • false: Allows URLs containing userinfo

• allow_subdomains (optional): Whether to allow subdomains of allowed domains.

  • false (default): Only exact domain matches (e.g., example.com allows example.com and www.example.com)
  • true: Allows subdomains (e.g., example.com allows api.example.com)

Mend Platform configuration

When a policy is loaded from the Mend Platform API the URL filter is identified by "id": "url". Its per-row config object uses camelCase keys that are mapped to the SDK's snake_case URLConfig fields as follows:

API field (config.*)	URLConfig field	Default
allowList	url_allow_list	[]
allowedSchemes	allowed_schemes	["https"]
customSchemes	merged into allowed_schemes	[]
blockUserinfo	block_userinfo	true
allowSubdomains	allow_subdomains	false

customSchemes lists non-standard schemes the operator wants to permit (e.g. myapp://). They are merged with allowedSchemes into a single deduplicated allowed_schemes set.

Example API response row:

{
    "id": "url",
    "action": "Alert",
    "status": true,
    "direction": "Output",
    "guardrail": "URL Filter",
    "config": {
        "allowList": ["company.com"],
        "allowedSchemes": ["https"],
        "customSchemes": [],
        "blockUserinfo": true,
        "allowSubdomains": false
    }
}

If config is absent the guardrail is loaded with all defaults (equivalent to URLConfig()).

Implementation Notes

• Detects URLs, domains, and IP addresses using regex patterns
• Validates URL schemes and security policies
• Supports exact domain matching or subdomain inclusion
• Handles IP addresses and CIDR ranges

What It Returns

Returns a GuardrailResult with the following info dictionary:

{
    "guardrail_name": "URL Filter",
    "config": {
        "allowed_schemes": ["https"],
        "block_userinfo": true,
        "allow_subdomains": false,
        "url_allow_list": ["example.com"]
    },
    "detected": ["https://example.com", "https://user:pass@malicious.com"],
    "allowed": ["https://example.com"],
    "blocked": ["https://user:pass@malicious.com"],
    "blocked_reasons": ["https://user:pass@malicious.com: Contains userinfo (potential credential injection)"]
}

Response Fields

• guardrail_name: Name of the guardrail that was executed
• config: Applied configuration including allow list, schemes, userinfo blocking, and subdomain settings
• detected: All URLs detected in the text using regex patterns
• allowed: URLs that passed all security checks and allow list validation
• blocked: URLs that were blocked due to security policies or allow list restrictions
• blocked_reasons: Detailed explanations for why each URL was blocked