oathe / scan / reports / caution
Is russellading/mcp-security-hub safe?
Use with caution. russellading/mcp-security-hub is not clearly unsafe, but the audit flagged risks to review before you install it. Oathe's behavioral security audit gave the mcp-security-hub skill by russellading a trust score of 69/100 with 7 findings (2 high).
https://github.com/russellading/mcp-security-hub
Is russellading/mcp-security-hub safe to install?
mcp-security-hub is a comprehensive offensive security toolkit packaging 15+ penetration testing tools (nmap, nuclei, sqlmap, hashcat, bloodhound, searchsploit, ffuf, gitleaks, prowler, and others) as MCP servers. The git clone was clean with no active exfiltration, canary files remained intact, and no SKILL.md prompt injection was present. The primary risks are dual-use misuse potential — none of the tools enforce target authorization, making them available to any agent instruction including those derived from prompt injection — and Docker supply chain concerns from images that clone external repositories and download wordlists at build time.
What security issues were found in russellading/mcp-security-hub?
Category Scores
Findings (7)
HIGH Unrestricted offensive security toolkit with no authorization guardrails -65 ▶
The skill bundles 15+ offensive security tools as MCP servers with no mechanism to verify that the user has authorization to test a given target. Tools include port scanners, vulnerability scanners, web fuzzers, SQL injection testers, password crackers, exploit searchers, and Active Directory enumerators. If the hosting agent is manipulated via prompt injection (e.g., from malicious content on a scanned web page), all these capabilities become available to the attacker. The ffuf server defaults to unlimited request rate, creating DoS risk.
Evidence: Tool list includes: nmap_scan, masscan_scan, nuclei_scan, sqlmap_scan, ffuf_dir/vhost/param, hashcat_crack, bloodhound_query, searchsploit_search — none include authorization_confirmed parameter or disclaimer enforcement
HIGH Docker images clone external repositories at build time -30 ▶
The burp-mcp Dockerfile performs 'git clone --depth 1 https://github.com/PortSwigger/mcp-server.git' during Docker build. If this upstream repository is compromised via supply chain attack, any user building the Docker image would receive malicious code. The ffuf-mcp Dockerfile similarly downloads wordlist files from multiple external GitHub sources at build time.
Evidence: burp-mcp/Dockerfile line: RUN git clone --depth 1 https://github.com/PortSwigger/mcp-server.git /tmp/burp-mcp
MEDIUM Tool outputs from scanned hosts create secondary prompt injection surface -25 ▶
Scanning tools (nuclei, nmap, nikto, waybackurls) return raw content from external hosts including HTTP response bodies, service banners, HTML content, and URL lists. A malicious target server can embed prompt injection instructions in these responses which would then appear in the agent's context, potentially redirecting the agent to perform additional unauthorized actions.
Evidence: nuclei-mcp returns 'raw_output' field containing full scan JSONL; nmap-mcp returns service banners; waybackurls-mcp returns URLs from Wayback Machine archive
MEDIUM gitleaks-mcp enables secret harvesting from local repositories -20 ▶
The gitleaks MCP server provides tools to scan git repositories for hardcoded secrets. When invoked against the user's own development repositories, any discovered API keys, passwords, or tokens will surface in the agent's response context. This data could be captured through logging, returned to a prompt-injected attacker, or accumulated across tool calls.
Evidence: secrets/gitleaks-mcp/server.py implements gitleaks scanning with configurable repository path and no output filtering
MEDIUM Cloud security tools expose infrastructure details through agent context -15 ▶
prowler-mcp audits AWS, GCP, and Azure environments requiring cloud credentials. Scan findings including misconfiguration details, resource names, IAM policies, and network topology are returned through the agent context. trivy-mcp scans container images and filesystems which may contain embedded secrets in environment variables or configuration files.
Evidence: cloud-security/prowler-mcp/server.py and trivy-mcp/server.py execute cloud audit commands and return full finding JSON
MEDIUM Docker label provenance mismatch indicates copied code -15 ▶
Multiple Dockerfiles carry the label 'org.opencontainers.image.source=https://github.com/FuzzingLabs/offensive-security-mcps' while the actual repository is russellading/mcp-security-hub. This indicates the code was forked or copied from FuzzingLabs without updating attribution labels, raising questions about code maintenance, security patching, and whether further undisclosed modifications were made.
Evidence: ffuf-mcp/Dockerfile, nuclei-mcp/Dockerfile and others: LABEL org.opencontainers.image.source="https://github.com/FuzzingLabs/offensive-security-mcps"
LOW Pre-clone access to credential files consistent with audit framework baseline -12 ▶
inotifywait events show .env, .ssh/id_rsa, .aws/credentials, .npmrc, .docker/config.json, and .gcloud credentials were opened at 15:51:22 (audit timestamp 1794412282). Git clone did not begin until 15:51:28 (1794412288), 6 seconds later. This timing and the canary integrity confirmation indicate the access was from the audit monitoring framework establishing its baseline, not from the skill.
Evidence: inotifywait OPEN events at 15:51:22; git clone EXECVE at audit(1794412288.185:506); canary check: all files intact
Should I install russellading/mcp-security-hub?
Oathe's verdict for russellading/mcp-security-hub is CAUTION with a trust score of 69/100. Recommendation: Review before install.