Writing code with AI is becoming increasingly convenient, but security issues are also becoming more concealed. The code you generate with tools like Cursor may contain SQL injections; your MCP configurations could be compromised; or malicious commands might be hidden in your CLAUDE.md files.
Ship Safe is designed to address this problem—it scans your entire project with just one command and identifies all potential security vulnerabilities.
What is it?
Ship Safe is an open-source AI-based security scanning tool. It uses 23 security agents working in parallel to detect over 80 types of attacks. There’s no need for registration or an API key, and it can be used offline.
npx ship-safeJust one command to start the scan.
What can it detect?
Code vulnerabilities:
- SQL/NoSQL injections, command injections, XSS
- JWT vulnerabilities (alg, weak keys)
- CSRF, OAuth configuration errors, TLS bypass
- SSRF (Server-Side Request Forgery)
AI/LLM-specific vulnerabilities:
- Prompt injections (OWASP LLM Top 10)
- Compromised MCP tools, agent hijacking
- RAG (Random Access Generation) context injections, document poisoning
- Exposure of system prompts
- Agent memory poisoning (to hide Unicode payloads)
Supply chain security:
- Obfuscated dependencies, typosquatting
- Suspicious installation scripts
- Errors in Docker/Terraform/K8s configurations
Other:
- Hardcoded secrets (including Git history)
- Compromises in CI/CD pipelines
- PII (Personally Identifiable Information) compliance checks
Let’s give it a try
I ran the tool on my own Astro blog project:
Security Score: 84.6/100 B — Minor issues to review
Category Breakdown
────────────────────────────────────────────────
✔ Secrets clean +0
✔ Code Vulnerabilities clean +0
✘ Dependencies 4 issue(s) -13 pts
✔ Auth & Access Control clean +0
✔ Configuration clean +0
✘ Supply Chain 2 issue(s) -2.4 pts
✔ API Security clean +0
✔ AI/LLM Security clean +0There were 4 CVE (Common Vulnerabilities and Exposures) related to dependencies and 2 supply chain issues. The code itself was fine.
The 23 security agents
Each agent is specialized in detecting a specific type of vulnerability and scans in parallel, automatically skipping irrelevant parts of the project:
| Agent | Detection Content |
|---|---|
| InjectionTester | SQL injections, command injections, XSS, path traversal |
| AuthBypassAgent | JWT vulnerabilities, CSRF, OAuth errors |
| SSRFProber | Server-Side Request Forgery |
| SupplyChainAudit | Obfuscated dependencies, suspicious installation scripts |
| ConfigAuditor | Docker/K8s/Terraform configurations |
| LLMRedTeam | Prompt injections, system prompt exposures |
| MCPSecurityAgent | MCP tool compromises, input validation |
| AgenticSecurityAgent | Agent hijacking, privilege escalation |
| RAGSecurityAgent | Context injections, document poisoning |
| MemoryPoisoningAgent | Agent memory poisoning |
| … | 13 more agents |
How to use it
# Interactive scan
npx ship-safe
# Complete audit
npx ship-safe audit .
# View the score only
npx ship-safe score .
# AI-based automatic fixes
npx ship-safe agent .
# Integration with CI/CD
npx ship-safe ci . --threshold 80Who is it for?
- Developers who write code with AI (users of tools like Cursor, Copilot, Claude Code)
- Teams that develop AI agents
- Projects that integrate MCP with external tools
- Open-source projects that need security audits
Project link: https://github.com/asamassekou10/ship-safe
It has 721 stars on GitHub and uses the MIT License. You can use it with just one command: npx ship-safe. No matter what AI tool you use for coding, running this scan is always a good idea.