"Using a search engine to find evidence to answer a question" matching MCP servers:

Audits AI agent skills for safety using static, semantic, adversarial, and supply-chain analysis, providing scores and risk flags. Can be run via CLI, CI, or as an MCP tool from Claude Code, Cursor, and Codex.

Enables generating falsifiable fleet hypotheses, compiling them into minimal bounded probe plans with control cohorts, and evaluating observations to return supported, refuted, or inconclusive verdicts.

Wraps Kali Linux penetration testing tools for authorized security assessments, driven by a local LLM via LM Studio.

Enables searching and analyzing CVEs and vulnerabilities from multiple sources, optimized for PR review scenarios to help developers identify the latest security issues.

Unified vulnerability search MCP server for penetration testing agents, integrating 5 data sources (NVD, OSV, EPSS, CISA KEV, Exploit-DB+GitHub) and 10 MCP tools for CVE query, keyword search, batch query, EPSS scoring, KEV checking, exploit search, and comprehensive assessment with Chinese output.

Passive website security and trust auditor that checks for security, SEO, AI surface, email, and other exposures, producing a score and remediation plan.

Enables agents to inspect web pages for Cloudflare Turnstile challenges without an API key, and clear Turnstile or WAF challenges using a SolveGate API key.

Connects AI assistants to Solodit's 49,000+ blockchain vulnerability database, enabling search, browse, and lookup of audit findings directly from your AI workflow.

Provides access to Atomic Red Team tests, enabling search, validation, and execution of atomic tests via natural language.

Enables searching and querying HackTricks pentesting documentation directly from Claude, with tools for quick lookup, grouped search results, page outlines, section extraction, and cheatsheet mode.

Enables prioritizing vulnerabilities by real-world exploitation evidence, combining CISA KEV, EPSS, CVSS, and SSVC to rank CVEs and package versions, and providing fix recommendations.

Enables AI assistants to search and analyze vulnerabilities and exploits from multiple intelligence sources, including NVD, CISA KEV, ExploitDB, Metasploit, and more, with tools for CVE research, exploit analysis, and report generation.

A read-only MCP server for the Qualys PCI Merchant API that lets LLM assistants answer questions about PCI compliance posture, such as which hosts are failing PCI or listing high findings.

Exports vulnerability and asset data from Rapid7 Command Platform via the Bulk Export API into a local DuckDB database, enabling AI assistants to run natural language queries and SQL analysis with sandboxed security.

Enables MCP-compatible agents to query security checklists, protocols, and framework mappings to safely build and operate AI agents, with tools for checklist runs, protocol lookup, mapping lookup, and threat search.

Agent-native "safe to ship?" security gate for AI-generated code. Uses real parsers and inter-rocedural taint analysis (JS/TS, Python, Go) to flag the classes AI coding agents get wrong — secrets, SQL injection, SS, SSRF, path traversal, command injection, weak JWT/CORS — and ranks findings by confidence. Exposes a scan tool over MCP.

Domain security reconnaissance for AI agents — 13 tools (DNS+DNSSEC, SSL/TLS, HTTP security headers, SPF/DKIM/DMARC email auth, port scan, ASN, RDAP/WHOIS) plus a one-shot security_scan returning a 0–100 Health Score (A–F). Free, no API key.

This server enables AI-assisted APK reverse-engineering entirely on-device, orchestrating jadx, apktool, adb, frida, and APKiD through a job/workflow engine, and exposing those agents as native MCP tools for Claude without any cloud dependency.

Open behavioral litmus for MCP servers — grades A–F across tool-output injection, egress, sensitive-data, and adversarial-input, with reproducible, content-addressed evidence. Tools: run_litmus, verify_attestation.

Connects MCP clients to pre-submission review that argues against a draft bug-bounty report or smart-contract finding the way a triager would, tying every claim to a supplied file and line and returning a submit, rewrite-then-submit, prove-first, hold-duplicate or drop verdict. Exposes tools to list review profiles, prepare reviews for the agent's own model, build hash-verified review packets, and — with a connection token — run hosted profiles on your own provider key.