Agentic DevSecOps Suite MCP server
Commit Oct 15, 2026Enables AI coding assistants to perform autonomous purple-team engineering, vulnerability validation, and dual-sided security remediation.
Also known as GABKYUd/agentic-devsecops-suite
- Stars
- 1
- Tools
- —
- Downloads, 30 days
- —
- Latest version
- —
- Last commit
- Oct 15, 2026
- License
- MIT
What is Agentic DevSecOps Suite MCP?
Agentic DevSecOps Suite enables AI coding assistants to perform autonomous purple-team engineering, vulnerability validation, and dual-sided security remediation.
Summary generated from the project's README
About Agentic DevSecOps Suite MCP server
Security engineers and developers using AI coding assistants like Cursor, Claude Code, or GitHub Copilot can use this suite to transform their assistants into defensive security partners. It provides a modular framework for autonomous purple-team engineering, moving beyond simple code audits to perform empirical vulnerability validation. By integrating this server, an AI assistant gains access to over 20 vulnerability classes and specialized tools for reconnaissance, hunting, and remediation.
The system operates through a five-phase methodology that includes attack surface mapping and deep vulnerability hunting. It distinguishes itself by enforcing a strict seven-question verification gate to prevent hallucinations and false positives. Instead of just flagging issues, the suite produces dual-sided outputs: a developer patch for application security and a Sigma or WAF detection rule for blue team operations. Users can connect the server via the Model Context Protocol to enable live tool access, including validation probes that can interface with Kali Linux environments via WSL2, Docker, or virtual machines.
What you can do with it
- 1Validate vulnerabilities with live probesAn AI assistant can generate and execute non-destructive test probes against a local environment to prove a finding is real.
- 2Generate dual-sided security outputsDevelopers can receive both a code fix and a corresponding detection rule for their security operations center.
- 3Automate attack surface mappingYou can instruct an AI to map out API endpoints and entry points to understand the reachable attack surface.
- 4Filter out AI hallucinationsThe framework uses a specific verification gate to ensure that theoretical vulnerabilities are actually reachable and impactful.
Written from the project's README. Every use case is checked against a quote from it.
Automated checks
Ran Nov 15, 2026Full report →Not verified yet. A server needs a score of 80, no serious finding, and enough data for us to check. How checks work · What to fix for more points
- Source repository linked and reachablegabkyud/agentic-devsecops-suite is public and reachablegithub.com/gabkyud/agentic-devsecops-suite ↗Pass
- Package names this repository as its sourceNot checked
- Build provenance (Sigstore attestation)Not checked
- Open source license (OSI)MITspdx.org/licenses/MIT.html ↗Pass
- Commit or release in the last yearLast commit or release 2026-10-15github.com/gabkyud/agentic-devsecops-suite ↗Pass
- OpenSSF ScorecardNot checked
- Package name not close to a popular packageNot checked
- No critical advisory in the package or its dependenciesNot checked
- No hidden instructions to the model0 tool descriptions and the README read; no override phrases, credential paths or hidden charactersPass
- Destructive tools are labelledNot checked
- Remote endpoint answers the MCP handshakeNot checked
README
View on GitHub🛡️ Agentic DevSecOps Suite
Autonomous Purple-Team Engineering & Vulnerability Validation Framework for AI Coding Assistants
🌌 Overview
The Agentic DevSecOps Suite is an open-source, modular purple-team engineering framework designed to transform AI coding assistants (Google Antigravity, Cursor, Claude Code, GitHub Copilot, Windsurf) into rigorous, empirically validated defensive security partners.
Instead of relying on generic chat prompts that hallucinate false positives or produce theoretical vulnerability noise, this suite enforces a 5-Phase Closed-Loop Methodology paired with The Strict 7-Question Verification Gate.
graph TD
A["Phase 1: Recon & Attack Surface Mapping"] --> B["Phase 2: Deep Vulnerability Hunting"]
B --> C{"Phase 3: The Strict 7-Question Gate"}
C -->|Fails Gate| D["DISMISSED: Hallucination / Neutralized"]
C -->|Passes Gate| E["Phase 4: Dual-Sided Output Synthesis"]
E --> F["Developer PR Patch (AppSec Fix)"]
E --> G["Blue Team Detection Rule (Sigma / WAF)"]
F --> H["Phase 5: Continuous Purple Team Validation (DAST / Harness)"]
G --> H
💡 How & Why This Suite Was Built
The Problem With Traditional AI Code Auditing
When standard Large Language Models (LLMs) are asked to "audit this codebase for security", they commonly suffer from two critical failures:
- Static AST Hallucinations & False Positives: LLMs flag raw string patterns as critical SQLi or XSS even when underlying frameworks (such as React JSX auto-escaping, ORM parameter bindings, or Cloudflare Workers) completely neutralize the issue in practice.
- One-Sided, Non-Actionable Criticisms: The assistant dumps a list of theoretical criticisms without providing tested root-cause code patches, automated regression tests, or telemetry detection rules for security operations.
The Engineering Solution
Developed at KYO INC during high-assurance application audits, this framework was created to establish a verifiable, zero-unverified-findings policy:
- Every finding must be validated in practice (via live DAST probes or empirical validation test harnesses) before being reported.
- Every accepted finding must produce dual-sided outputs: an architectural code fix for the software engineer AND a Sigma / WAF detection rule for the blue team.
🎁 A Free, Modular Template: Adopt What Matters to You
[!NOTE] This repository is a free, MIT-licensed template. You do not need to adopt the entire suite at once. Modular by design, you can pick and choose only the specific skills and playbooks relevant to your stack and discard the rest.
🧭 Adoption Guide: What to Pick for Your Project
| Your Project Focus | What to Adopt | What You Can Skip |
|---|---|---|
| Web Apps & REST/GraphQL APIs | • bug-hunting-recon-surface• vuln-hunting-auth-access-control• vuln-hunting-injection-rce |
Skip Active Directory & Cloud Infrastructure playbooks. |
| CI/CD & Supply Chain Security | • devsecops-pipeline-automation.md• bug-bounty-validator-gate• threat-intel-skills-distiller |
Skip application-specific routing heuristics. |
| SOC & Incident Response Readiness | • incident-response-playbooks• detection-rules-starter-reference.md• devsecops-defense-playbooks.md |
Skip static source code reconnaissance. |
| AI Assistants with MCP Support | • .agents/mcp/server.js• devsecops-mcp-kb |
Run the MCP server to give your LLM live tool access to 20+ vulnerability classes. |
⚖️ What to Adopt vs. What to Avoid
✅ Best Practices to Adopt:
- Enforce the 7-Question Gate: Never file a security ticket or commit a patch based solely on static code patterns. Ensure direct reachability, identify access prerequisites, and prove business impact.
- Skill Transparency & Developer Approval Gate: Autonomous agents must never silently or unilaterally activate audit skills or trigger inspection routines in the background. Before any skill is selected or executed, the agent must clearly explain to the developer why the skill is needed, what files or routes it will analyze, and what probes/tools it will run — requiring explicit developer sign-off (
Proceed / Reject) before taking action. - Dual-Sided Output: Always pair an AppSec code patch with a corresponding Blue Team detection rule (Sigma / log query / WAF signature).
- Strict Remote Push Gate: Never permit automated agents to push to remote git repositories without prior explicit user approval.
- Defense-in-Depth Defaults: Always mandate baseline security headers (
Helmet/CSP), CORS origin allowlists, and rate-limiting on unauthenticated endpoints.
🚫 Anti-Patterns to Avoid:
- Avoid Offensive-Only Weaponization: Do not use AI agents to generate novel functional exploits or weaponized malware payloads against targets. Keep the scope strictly defensive, remediative, and detection-focused.
- Avoid Monolithic Prompt Dumps: Don't dump 50 pages of instructions into every prompt. Use specialized, on-demand skills (
SKILL.md) that activate only when auditing relevant subsystems. - Avoid Hardcoded Secrets: Never hardcode API keys, production tokens, or private credentials in agent configuration files or git history.
📁 Repository Structure
agentic-devsecops-suite/
├── .agents/
│ ├── AGENTS.md # Core threat modeling, MITRE ATT&CK & defensive rules
│ ├── INDEX.md # Master reference sitemap & documentation index
│ ├── purple-team-vuln-class-matrix.md # Comprehensive 20+ vulnerability class purple matrix
│ ├── purple-team-automation-methodology.md # The 5-Phase operational methodology
│ ├── devsecops-pipeline-automation.md # GitHub Actions CI/CD security gate blueprints
│ ├── devsecops-defense-playbooks.md # Blue team defense & hardening guides
│ ├── detection-rules-starter-reference.md # Baseline Sigma detection rules
│ ├── playbook-*.md # 7 Dedicated Incident Response Playbooks
│ │ ├── playbook-ad-compromise-response.md
│ │ ├── playbook-c2-beacon-response.md
│ │ ├── playbook-cloud-credential-compromise.md
│ │ ├── playbook-container-kubernetes-compromise.md
│ │ ├── playbook-phishing-response.md
│ │ ├── playbook-ransomware-response.md
│ │ └── playbook-web-app-exploitation.md
│ ├── mcp/
│ │ └── server.js # DevSecOps Knowledge Base MCP Server
│ ├── mcp_config.json # Portable Model Context Protocol configuration
│ └── skills/ # Reusable Agent Skills (SKILL.md specs)
│ ├── bug-hunting-orchestrator/ # Master 5-phase orchestrator
│ ├── bug-hunting-recon-surface/ # Route & entry point discovery
│ ├── bug-bounty-validator-gate/ # 7-Question Gate triage filter
│ ├── vuln-hunting-auth-access-control/ # IDOR, JWT, OAuth, Mass Assignment
│ ├── vuln-hunting-injection-rce/ # SQLi, NoSQLi, Command Injection, SSTI
│ ├── vuln-hunting-network-ssrf-logic/ # SSRF, Path Traversal, CORS, TOCTOU
│ ├── threat-intel-skills-distiller/ # Ingest threat intel into Semgrep/Sigma
│ ├── incident-response-playbooks/ # Operational triage workflows
│ └── devsecops-mcp-kb/ # MCP server interaction skill
├── .gitattributes # Normalized LF line endings
├── .gitignore # Clean exclusion rules
├── CONTRIBUTING.md # Open-source contribution guidelines
├── LICENSE # MIT License
├── package.json # MCP executable package definition
├── README.md # You are here
└── SECURITY.md # Vulnerability reporting & security policy
🚀 Quickstart: Integrating Into Your Project
1. Drop Into Your Repository
Copy the .agents/ folder into the root of your existing codebase:
cp -r .agents /path/to/your/project/
2. Configure Your AI Assistant
- Antigravity / Cursor: Point your agent customizations or workspace rules to
.agents/AGENTS.md. The skills in.agents/skills/*/SKILL.mdwill be discovered and loaded dynamically on demand. - Model Context Protocol (MCP): If your assistant supports MCP, register the DevSecOps knowledge server in your editor's MCP configuration. The MCP server exposes tools for vulnerability lookups, Kali DAST validation probes, 7-Question Gate evaluations, and the Skill Approval Gate (
explain_and_request_skill_approval), ensuring your agent transparently explains and requests approval before executing any audit skill:{ "mcpServers": { "devsecops-knowledge": { "command": "node", "args": ["./.agents/mcp/server.js"] } } }
3. Prompting Your Assistant
You can invoke the workflow directly in conversation:
"Run the bug-hunting-orchestrator across our API endpoints to map our attack surface, inspect access controls, and evaluate any findings through the 7-Question Gate."
🐉 Validating Findings with Kali Linux (WSL2, Docker & Virtual Machines)
[!WARNING]
⚠️ Rules of Engagement & Authorization Notice
Only run network scanners, fuzzers, and dynamic probes (
nmap,ffuf,nuclei,curl, etc.) against systems, networks, and applications that you own or for which you have explicit, written authorization to test.While this suite is provided under the permissive MIT license, running scanners or active probes against third-party systems without prior authorization violates terms of service, acceptable use policies, and computer crime laws (such as the US CFAA or Brazilian Law 12.737/2012). All verification probes and tools in this framework are engineered strictly for first-party defensive testing on local loopback (
localhost,127.0.0.1) and authorized private staging environments.
A core tenet of this framework is Mandatory Empirical Validation:
Never file a security bug or commit a patch based solely on static code pattern assumptions. Every finding must be empirically proven to trigger against a running environment, and every patch must be proven to neutralize it.
By pairing the DevSecOps MCP Server (get_wsl_validation_probe) with Kali Linux, your AI assistant bridges the gap between static AST analysis (SAST) and live runtime verification (DAST). The assistant can generate targeted, non-destructive test probes, execute them against your local/staging environment, and feed the live HTTP responses directly into the 7-Question Gate.
graph LR
subgraph Host["Host Workspace"]
SAST["AI Coding Assistant<br>(SAST / Code Audit)"]
Target["Target Application<br>(localhost / Staging)"]
end
subgraph MCP["DevSecOps MCP Server"]
ProbeGen["get_wsl_validation_probe"]
Gate["evaluate_7_question_gate"]
end
subgraph Kali["Kali Linux Engine (DAST)"]
WSL["Option 1: WSL2 Instance<br>(Near-native local speed)"]
Docker["Option 2: Docker Container<br>(Ephemeral & CI/CD ready)"]
VM["Option 3: Dedicated VM<br>(Full network isolation)"]
end
SAST -->|"1. Request Probe"| ProbeGen
ProbeGen -->|"2. Synthesize Command"| SAST
SAST -->|"3. Execute Probe"| WSL & Docker & VM
WSL & Docker & VM -->|"4. Send Benign Payload"| Target
Target -->|"5. Return Status & Headers"| WSL & Docker & VM
WSL & Docker & VM -->|"6. Real Telemetry"| SAST
SAST -->|"7. Evaluate Findings"| Gate
Gate -->|"8. Verdict: PASS"| Fix["Developer PR Patch + Sigma Rule"]
Gate -->|"8. Verdict: DISMISSED"| Trash["Dismissed"]
🛠️ Execution Options: WSL2 vs. Docker vs. VMs
You can run Kali Linux alongside this suite using whichever virtualization model fits your development and infrastructure workflow:
Option 1: WSL2 (Windows Subsystem for Linux) — Recommended for Local Dev
WSL2 provides near-zero overhead, boots in milliseconds, and shares localhost networking directly with Windows. Probes running inside Kali can hit your local dev servers (http://localhost:3000) without manual port forwarding.
- Install Kali on WSL2:
wsl --install -d kali-linux - Install Baseline Validation Tools:
sudo apt update && sudo apt install -y curl nuclei ffuf nmap - Execution via AI Assistant / PowerShell:
# Non-destructive SQLi check wsl -d kali-linux -e curl -s -o /dev/null -w "%{http_code}" "http://localhost:3000/api/items?id=1%27%20OR%201=1--" # Non-destructive CORS reflection check wsl -d kali-linux -e curl -s -I -H "Origin: https://evil.attacker.com" "http://localhost:3000/api/user"
Option 2: Docker Containers (kalilinux/kali-rolling) — Disposable & CI/CD Ready
If you prefer a sandboxed environment that leaves no persistent files on your host system, or if you run on Linux, macOS, or automated CI/CD runners (GitHub Actions / GitLab CI), Docker is ideal.
- Ephemeral Interactive Shell:
docker run --rm -it --net=host kalilinux/kali-rolling /bin/bash - One-Liner DAST Probes:
# Execute curl probe using host networking docker run --rm --net=host kalilinux/kali-rolling curl -i -s "http://127.0.0.1:3000/api/orders/1" # Run Nuclei vulnerability scan from Docker docker run --rm --net=host projectdiscovery/nuclei -u "http://127.0.0.1:3000" -tags misconfig,cve
Option 3: Dedicated Virtual Machines (VirtualBox / VMware / Proxmox) — Enterprise & AD Labs
When performing complex purple-team simulations involving Active Directory, internal network pivoting, or strict hypervisor-level network segregation, run a standalone Kali Linux VM.
- Pre-built VMs: Download official images from kali.org/get-kali/#kali-virtual-machines.
- Network Setup: Attach a Host-Only or Internal Virtual Network adapter to isolate the testing subnet from production internet egress.
- Automation: Trigger probes remotely over SSH from your agent or CI pipeline:
ssh kali@<KALI_VM_IP> "curl -i -s 'http://<TARGET_HOST>:8080/api/auth/profile'"
🔄 The Closed-Loop Verification Workflow
- Suspect Code Identified & Gate Evaluated: The AI assistant flags a potential security flaw (e.g. an unvalidated IDOR parameter in an API controller) and runs
evaluate_7_question_gate. When the gate passes, the server issues a cryptographically secureapproval_token. - DAST Probe Synthesized: The assistant calls the MCP server tool with the gate's approval token:
Gate Enforcement & Target Guard:{ "name": "get_wsl_validation_probe", "arguments": { "approval_token": "<APPROVAL_TOKEN_FROM_GATE>", "vuln_class": "idor", "target_url": "http://localhost:3000/api/orders/1", "auth_header": "Bearer <TOKEN_USER_B>" } }get_wsl_validation_proberefuses execution ifapproval_tokenis omitted or invalid. Its built-in target guard restricts destination URLs to localhost and private ranges, preventing SSRF or steering toward third-party hosts. - Runtime Execution: The assistant runs the generated command via WSL2, Docker, or VM against the running dev server.
- Empirically Validated Analysis (Differential Testing):
- Important: An HTTP 200 alone does not prove an IDOR (shared or public data legitimately returns 200). Empirical validation requires differential verification: testing whether User B's token retrieves User A's private, non-shared records while unauthenticated requests are denied.
- If User B successfully receives User A's unshared private data, the flaw is empirically proven.
- If the framework middleware returns
HTTP 403 ForbiddenorHTTP 404 Not Found, the finding is immediately dismissed as a false positive.
- Remediation & Regression Check: The assistant writes the AppSec patch, restarts the test server, and re-runs the exact same Kali probe to verify that the vulnerability is neutralized.
📜 License & Attribution
This project is licensed under the MIT License — you are free to modify, distribute, and embed this template into personal, open-source, or commercial projects.
Maintained with 🛡️ by Gabriel Kyo & KYO INC. Contributions, issues, and discussions are welcome!
Agentic DevSecOps Suite MCP questions
How does the framework prevent AI hallucinations in security audits?
It uses a 5-Phase Closed-Loop Methodology and a Strict 7-Question Verification Gate to ensure findings are empirically validated before reporting.
Can I use this with my existing AI coding assistant?
Yes, it is designed for assistants like Cursor, Claude Code, and GitHub Copilot, and can be integrated via the Model Context Protocol (MCP).
Is the suite modular?
Yes, it is a modular template that allows you to pick specific skills and playbooks relevant to your stack while skipping others.