Neura Market
pentest-ai-agents logo

pentest-ai-agents

Free

Turn Claude Code into your offensive security research assistant. Specialized AI subagents for authorized penetration testing plan engagements, analyze recon, research exploits, build detections, audi

CybersecurityFreeFree tier
Type
Open Source
Visit Website GitHub (805 stars)

About pentest-ai-agents

ptai is an open-source AI penetration testing tool that runs locally as a CLI and MCP server. It integrates with Claude Code (or Cursor) to act as offensive security agents that plan engagements, analyze reconnaissance data, research exploits, build detections, audit STIGs, and write reports. Findings are structurally verified: a candidate claim becomes VERIFIED only when a named machine oracle re-runs the exploit and reproduces it three-for-three. The tool supports 14 vulnerability classes including SQL injection, stored/reflected XSS, SSRF, XXE, BOLA/IDOR, JWT alg:none, and more. Every verified finding ships as a portable proof capsule that can be replayed by anyone. It outputs SARIF for CI integration (e.g., GitHub Code Scanning) and can gate builds on verified findings only. Runs entirely on the user's machine with local SQLite storage, no telemetry, and no data exfiltration. Can operate without any API key (deterministic mode) or bring your own LLM key.

Key Features

Local CLI and MCP server, runs entirely on your machine
Machine oracle verification: findings earn VERIFIED only when reproduced N-for-N by a named oracle
Portable proof capsules that can be replayed by anyone to reproduce findings
Integration with Claude Code and Cursor via MCP (47 tools, no API key required)
SARIF output for CI/CD gates (GitHub Code Scanning, --fail-on verified)
Quarantine mode: third-party scanner noise (nuclei, nikto, etc.) remains hidden until a ptai oracle independently reproves it
Attack chain verification: multi-step paths verified hop-by-hop with machine receipts
Supports 21 machine oracles covering 14 vulnerability classes (SQL injection, XSS, SSRF, XXE, IDOR, JWT alg:none, path traversal, open redirect, type confusion, etc.)
No telemetry, no hosted service; data stays in local SQLite
Deterministic mode available (--no-llm) for fully key-free operation

Pros & Cons

Pros
  • Machine-verified findings eliminate false positives; every badge is backed by reproducible proof
  • Fully local operation with no data exfiltration or telemetry
  • Portable proof capsules enable trust by reproduction, not screenshots
  • Seamless integration with Claude Code and Cursor for AI-driven attack planning
  • Wide coverage of common web vulnerability classes (14 types)
  • Can run without any API key in deterministic mode
  • CI-ready with SARIF and build-failing on verified findings only
Cons
  • Requires technical expertise to install and configure (CLI and MCP setup)
  • Currently supports only 14 vulnerability classes; not exhaustive
  • AI-driven attack planning depends on an external LLM (Claude subscription or bring-your-own)
  • Primarily designed for authorized penetration testing; not for casual security scanning
  • Limited out-of-the-box integration with other security tools beyond SARIF and MCP

Best For

Authorized penetration testing with machine-verified resultsContinuous security scanning in CI/CD pipelinesAutomated vulnerability verification and triageAuditing STIGs and compliance checksAnalyzing reconnaissance data and researching exploitsBuilding detections and writing security reports

Alternatives to pentest-ai-agents

Verkada

Advanced and User-Friendly Security Cameras by Verkada

SecureGPT

Build and Secure Your ChatGPT Plugins with SecureGPT by Escape

Fraud.net

Harness AI & Machine Learning for Superior Fraud Detection

MaskmyPrompt

Use MaskMyPrompt to anonymize your ChatGPT prompts

Final Touch

Enhance Your Browsing Experience with Customized Cookie Preferences

Skoot

Skoothere.com: Tailor Your Cookie Consent Preferences

FAQ

How does ptai verify findings?
A finding earns VERIFIED only when a named machine oracle re-runs the exploit and reproduces it N-for-N (typically 3 times). The code rejects any 'verified' verdict that cannot name the oracle that earned it. No LLM assertion or title-based shortcut is accepted.
What vulnerability classes does ptai test?
ptai supports 14 vulnerability classes including SQL injection, stored/reflected XSS, SSRF, XXE, BOLA/IDOR, JWT alg:none, path traversal, open redirect, type confusion, mass assignment, trusted-header injection, host-header poison, and sequential IDOR. There are 21 machine oracles covering these classes.
Does ptai require an API key or internet connection?
No. ptai runs entirely locally with no hosted service. You can use it with a Claude subscription (which requires internet for the LLM) or bring your own LLM key. It also supports a deterministic mode (--no-llm) that requires no API key at all.
How can I integrate ptai into my CI/CD pipeline?
ptai outputs SARIF formatted findings, suitable for GitHub Code Scanning. The bundled GitHub Action supports a --fail-on verified flag that breaks the build on proven findings only, never on unverified noise.