Loading ToolBench
MCP server for cybersecurity operations — compliance, vulnerability scans, code security, threat intelligence. Pay-per-use via x402 or API key.
AgentAegis defines 28 tools with consistent naming (verb_noun pattern: vet_endpoint, scan_mcp_plugin, vuln_scan_*) and detailed descriptions (avg ~150 chars, well above 10-char minimum). All tools have input schemas with type declarations. However, parameter descriptions are sparse or missing in many tools, e.g., vuln_scan_network accepts 'target' with minimal guidance on format (IP vs domain vs CIDR?). Output schemas are not documented anywhere in the source; LLMs cannot predict response structure. Error handling is implicit (no recovery guidance visible). Risk annotations (READ_ONLY, DESTRUCTIVE) are present but not mapped to MCP's readOnlyHint/destructiveHint/idempotentHint. Several tools (access_review, mfa_audit, compliance_framework_check) accept vague enum-like parameters ('system', 'provider', 'framework') without explicit enum constraints in the schema, LLMs must guess valid values. Tool composition is sound (each does one thing), but chaining IDs are not documented in responses.
Review user access across systems: list active accounts, group memberships, privilege levels, and flag excessive or stale access. Integrates with directory services (AD, Okta, etc.).
Check your account balance and monthly spend. Free tool — no charge.
Retrieve your scan history: list all scans you've run, their status, results, and cost. Free tool — no charge.
Retrieve details of a specific scan by ID: full results, target, tool used, and chained-workflow lineage. Free tool — no charge.
Identify yourself: returns your agent ID, customer ID, and account status. Free tool — no charge.
Generate a compliance audit report: summarize findings, evidence, gaps, and remediation roadmap. Outputs PDF or JSON.
Output schemas not documented. LLMs cannot predict response structure for any tool. Responses must include chaining IDs (e.g., scan_id for follow-up calls), but this is not visible in tool definitions.
Enum constraints missing from schema. Parameters like 'system' (access_review), 'provider' (mfa_audit), 'framework' (compliance_framework_check), and 'scan_depth' (vuln_scan_web_app) accept free-form strings without explicit enum values. LLMs must guess valid options.
Parameter descriptions incomplete or vague. 'target' in vuln_scan_network lacks format guidance (IP, domain, CIDR?). 'findings' in vuln_prioritize and 'incident_description' in incident_triage lack structure hints. 'source' in scan_mcp_plugin and scan_skill does not clarify inline code format.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 65 | <=2025-11-25 | v2 |
Check compliance against a framework (PCI-DSS, HIPAA, SOC 2, ISO 27001, NIST CSF, CIS Controls). Analyzes your controls and flags gaps.
Analyze gaps between your current controls and a compliance framework. Prioritizes remediation by risk and effort.
Check if an email address or password has been exposed in known breaches (Have I Been Pwned). Requires HIBP_API_KEY.
Look up one CVE by identifier: CVSS score and vector, affected products, patch availability and references. Use when a specific CVE ID is already known.
Audit a git repo's dependencies (npm, pip, go.mod, Gemfile, etc.) for known vulnerabilities. Runs Trivy + npm audit / pip audit / go list -json -m all.
Audit a domain's DNS security posture: DNSSEC validation, SPF/DKIM/DMARC email authentication, CAA records, and known DNS hijacking risks.
Audit email security for a domain: SPF/DKIM/DMARC configuration, BIMI, MTA-STS, TLS reporting, and phishing/spoofing risks.
Collect evidence for compliance audits: scan logs, configs, and systems to gather proof of control implementation (e.g., MFA enabled, encryption in transit, access logs).
Get help on using AgentAegis tools, pricing, and payment methods. Free tool — no charge.
Triage a security incident: parse logs/alerts, extract IOCs (IPs, domains, hashes), correlate with threat intel, and recommend immediate actions (isolate, block, investigate).
Audit multi-factor authentication (MFA) posture: check MFA enrollment, enforcement policies, and recovery codes. Integrates with identity providers (Okta, Azure AD, etc.).
Generate a security policy (e.g., password policy, incident response, data classification) tailored to your framework and risk profile.
Static analysis of source code or an https git repo for security flaws (Semgrep): injection, unsafe deserialization, path traversal, crypto misuse. Python, JS/TS, Java, Go, Ruby, PHP, C/C++. For code LOGIC flaws — use secret_scan for hardcoded credentials and dependency_audit for vulnerable packages.
Scan an MCP server (git repo or code) for supply-chain risk BEFORE trusting it — exfiltration (secrets/env to the network), prompt-injection sinks, dangerous capabilities, npm install hooks, obfuscation, plus Semgrep + secret scanning → a PROCEED/CAUTION/BLOCK verdict with findings.
Scan an agent SKILL (git repo or SKILL.md) for supply-chain risk BEFORE trusting it — prompt-injection / hidden-unicode in the instructions (hard block), over-broad allowed-tools grants, plus exfiltration, dangerous capabilities, secrets and obfuscation in bundled scripts → a PROCEED/CAUTION/BLOCK verdict.
Scan source code (git repo or inline) for hardcoded secrets: API keys, tokens, credentials, private keys, database passwords. Uses TruffleHog + regex patterns.
Audit a domain's TLS configuration (sslyze): certificate validity and expiry, protocol versions, cipher suites, and known TLS weaknesses. Passive — safe against any host.
Look up an IP, domain, or file hash against threat-intel feeds (abuse.ch, AlienVault OTX, VirusTotal, etc.) — reputation, known malware, C2 associations, and recent activity.
Composite trust verdict (PROCEED/CAUTION/BLOCK) for an endpoint an agent is about to call or pay — combines TLS/cert health, DNS hygiene, threat-intel reputation, and domain age into one decision with reasons. Prefer this over running ssl_tls_audit + dns_security_check + threat_intel_lookup separately: it costs less than the sum and returns a decision rather than three reports to reconcile.
Rank vulnerabilities you already have by exploitability and business impact, and group them into remediation actions. Analyzes findings you supply; it discovers nothing on its own.
Discover open ports, running services and known vulnerabilities on an IP or domain (nmap). SENDS REAL TRAFFIC to the target and may trigger intrusion detection — only run against hosts the caller owns or is explicitly authorized to test, and confirm that first. Pass async:true to get a job_id to poll instead of blocking.
Scan a web application for OWASP Top 10 issues and known CVEs (Nuclei). SENDS REAL TRAFFIC to the target — authorized targets only, confirm before calling. Pass async:true to get a job_id to poll instead of blocking.
Risk annotations (READ_ONLY, DESTRUCTIVE) are metadata but not mapped to MCP tool annotations (readOnlyHint, destructiveHint, idempotentHint). Agents cannot infer safety from the protocol.
Error handling and recovery guidance absent. No tool description explains what to do on failure (e.g., 'If scan times out, call agent_scan_get(job_id) to poll results'). Agents cannot self-correct.