Loading ToolBench
OSINT Agent Skills — 24 MCP tools for autonomous OSINT investigation. DNS, WHOIS, CT logs, Shodan, breach data, geolocation, crypto tracing. Works with Claude Code, Cursor, any MCP client.
This server exposes 23 OSINT tools with READ_ONLY risk classification. While tool names follow verb_noun patterns (dns_lookup, shodan_search, etc.), the source code snippet is truncated at tools/mcp-server.js line ~60, preventing verification of actual schema definitions, parameter types, output documentation, and error handling. Based on the visible package.json and tool registry (mcp-tools.json reference), descriptions appear present but are generic and lack actionable guidance. No evidence of input validation schemas, output structure documentation, pagination support, or error recovery patterns. The tools target external APIs (Shodan, VirusTotal, Hunter, HIBP, Etherscan, SecurityTrails, GitHub) but parameter handling for API keys via optional fallback to environment variables suggests potential secret leakage risk. The descriptions provided are between 50-90 characters, within the 10-1024 range but on the shorter side, lacking WHEN/WHY context.
Search Certificate Transparency logs for SSL/TLS certificates
Lookup CNAME records for DNS aliasing discovery
Trace cryptocurrency addresses and blockchain transaction flows
Monitor darkweb marketplaces for threat intelligence and data breaches
DNS lookup and resolution tool for hostname to IP address mapping
Lookup blockchain addresses and transaction data on Ethereum using Etherscan
Lookup geolocation information for IP addresses
Search GitHub for user profiles and code repositories
Source code truncated, cannot verify actual tool registration, input schema definitions, output documentation, or error handling implementation
API keys exposed as optional tool parameters (api_key fields in shodan_search, virustotal_scan, hunter_email_search, haveibeenpwned_search, etherscan_address_lookup, securitytrails_search), violates secret-injection pattern. Even with fallback to env vars, accepting keys as params risks logging and tracing secrets.
No input validation schemas visible. Parameters lack type constraints, enums, format validators, or length/range limits. E.g., 'hostname' in dns_lookup has no format or length constraints; 'query' in shodan_search has no examples or enum of valid search operators.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 35 | 2026-07-28+ | v2 |
Check if email addresses or passwords have been compromised in known data breaches
Search for email addresses associated with a domain using Hunter.io
Analyze malware samples and get threat intelligence reports
Lookup MX records for email infrastructure discovery
Lookup nameserver records for infrastructure mapping
Search Pastebin for leaked data and sensitive information disclosures
Perform reverse DNS lookup to find hostnames associated with IP addresses
Search DNS and WHOIS historical records using SecurityTrails
Search Shodan database for internet-connected devices and services
Search for social media accounts and presence related to an entity
Lookup SPF and TXT DNS records for email authentication and domain configuration
Enumerate subdomains for a given domain using DNS and public records
Build threat actor profiles from OSINT data and intelligence sources
Scan URLs, files, and IP addresses using VirusTotal threat intelligence database
Query WHOIS database for domain registration and contact information
Tool descriptions are generic, 50-90 characters, lacking WHEN/WHY context. E.g., 'Search Shodan database for internet-connected devices and services' does not explain when to use vs other search tools, expected output structure, or required prerequisites (e.g., Shodan API key availability).
Output schemas not documented. No indication of what fields are returned, data types, structure (flat vs nested), pagination support, or required downstream IDs (e.g., do search results include IDs for follow-up queries?).
No evidence of error handling or recovery guidance. If Shodan API is down, rate-limited, or returns no results, the LLM has no recovery path or alternative actions suggested.
No pagination or result limiting visible. Tools like 'pastebin_search' and 'darkweb_marketplace_monitor' may return hundreds of results, risking context window exhaustion and token waste.
No audit logging or permission gating visible. Tools accessing sensitive data (HIBP breach data, darkweb marketplaces, threat actors) should enforce access controls and log who called what with what parameters.
Tool naming uses underscores and is verb_noun, which is good; however, some names are ambiguous. E.g., 'social_media_footprint' does not clarify whether it returns accounts, profiles, or metadata. 'threat_actor_profile' does not explain if it returns attributes, relationships, or a structured threat model.