One Sentence In. Full Security Audit Out.
WAFtester ships 7 guided prompts that structure multi-step workflows. Say what you want tested. The AI handles every phase.
What "Autonomous" Means Here
You Give a Target
"Audit the WAF on staging.example.com" or "Find XSS bypasses on our API." One sentence. No flags, no config.
AI Picks the Tools
The prompt structures the workflow into phases. The AI selects from 27 MCP tools, chains them, and handles errors.
You Get Results
Detection rates, bypass chains, F1 scores, remediation guidance. Explained in natural language, not raw JSON.
Example: Full Security Audit
This is what happens when you use the security_audit prompt. One input. Five phases. Zero manual steps after the first sentence.
"Run a full security audit on https://staging.example.com. Focus on the API endpoints. We're behind Cloudflare."
security_audit: 5-phase autonomous workflow
Recon
AI calls probe and detect_waf. Gets TLS version, server headers, confirms Cloudflare with 98% confidence. Notes cf-ray header and 1020 error page as evidence.
Discovery
AI calls discover. Crawls the site, analyzes JavaScript, checks Wayback Machine. Finds 47 endpoints including 12 API routes under /api/v2/.
Test Plan
AI calls learn. Maps endpoints to OWASP categories, prioritizes /api/v2/users (auth endpoint, high risk) and /api/v2/search (user input, SQLi target). Generates attack matrix.
Scan
AI calls scan with Cloudflare-optimized tampers (charunicodeencode, randomcase). Tests 2,800+ payloads. Finds 7 bypasses: 5 XSS reflected, 2 SQLi union-based.
Assess and Report
AI calls assess. Computes detection rate (94.2%), F1 score (0.969), MCC (0.942). Assigns grade B+. Recommends tightening regex on "api-protection" rule group. Generates executive summary.
Total time: ~3 minutes. Total human input: one sentence. The AI explains every finding, suggests remediation, and can export results to SARIF, JSON, or HTML.
7 Guided Prompts
Each prompt structures a multi-step workflow. The AI follows the phases automatically. You can use them from Claude Desktop, VS Code Copilot, Cursor, or any MCP client.
security_audit
5-phase end-to-end security audit
Recon → Discovery → Test Plan → Scan → Assess. Handles everything from WAF fingerprinting to grade assignment.
Required: target (URL)
Optional: scope (full | api | sqli,xss,ssrf) · environment (production | staging | development)
"Run a security audit on https://staging.example.com, focus on API endpoints"
waf_bypass
6-step bypass discovery workflow
Detect vendor → select tampers → mutation matrix → validate chains → rank by severity → suggest WAF rule fixes.
Required: target (URL) · category (sqli | xss | ...)
Optional: stealth (true | false)
"Find XSS bypasses on https://app.example.com in stealth mode"
full_assessment
Enterprise assessment with compliance mapping
Quantitative scoring (F1, MCC, FPR) with OWASP Top 10 and PCI-DSS compliance mapping. Executive summary generation.
Required: target (URL)
Optional: compliance (owasp | pci-dss | both)
"Run an enterprise assessment on production with OWASP and PCI-DSS mapping"
discovery_workflow
Attack surface mapping
Crawl → JavaScript analysis → Wayback Machine → sitemap → generate a prioritized test plan with endpoint risk scoring.
Required: target (URL)
Optional: depth (shallow | normal | deep) · service (authentik | n8n | immich | ...)
"Map the attack surface of https://app.example.com with deep discovery"
evasion_research
Systematic tamper and mutator research
Tests 90+ tamper scripts against 49 mutator functions to find novel bypass chains. Ranks results by WAF-specific effectiveness.
Required: target (URL) · payload (base payload to test)
"Research evasion techniques for <script>alert(1)</script> against https://target.com"
template_scan
Nuclei template-based testing
Load Nuclei-compatible YAML templates (HTTP, DNS, TCP, UDP). Test WAF coverage against known CVEs and standard exploit patterns.
Required: target (URL)
Optional: focus (bypass | detection | all)
"Scan https://target.com with bypass-focused templates"
spec_security_audit
API spec-driven security audit
Feed an OpenAPI, Swagger, Postman, or HAR spec. The AI validates, extracts endpoints, runs intelligence analysis, and scans every route.
Required: spec_content (full spec YAML/JSON)
Optional: target (base URL override) · intensity (quick | normal | deep | paranoid)
"Audit the API defined in this OpenAPI spec at paranoid intensity"
What It Looks Like in Practice
Real conversations. The AI decides which tools to call, chains them, and explains the results.
Quick Bypass Hunt
You
"Can you find SQLi bypasses on our staging WAF? It's at https://staging.acme.com"
AI Agent
I'll start by identifying the WAF vendor, then run targeted bypass testing.
detect_waf → AWS WAF v2 (92% confidence)
list_tampers → 14 tampers effective for AWS WAF
scan → categories: sqli | tampers: between,equaltolike,space2comment
→ 3 bypasses found on /api/search endpoint
Found 3 SQL injection bypasses on the /api/search endpoint. The between tamper rewrites AND/OR keywords to bypass the string-match rule. I'd recommend adding a normalized-keyword rule to your AWS WAF rule group.
Pre-Deployment WAF Check
You
"We just updated our Cloudflare WAF rules. Run a full assessment and tell me if we're ready for production."
AI Agent
I'll run a complete assessment with false positive testing to verify the rules are both effective and not over-blocking.
detect_waf → Cloudflare (98% confidence)
scan → 2,800+ payloads across all categories
assess → with false positive corpus (Leipzig)
→ Grade: A | Detection: 97.1% | FPR: 0.1% | F1: 0.985
Your WAF is production-ready. Grade A with 97.1% detection rate and only 0.1% false positive rate. The only gaps are 2 niche SSTI payloads using double-encoded Unicode. Low risk for most applications, but I can show you the exact payloads if you want to add coverage.
API Spec Audit
You
"Here's our OpenAPI spec. Test every endpoint against our WAF." [attaches openapi.yaml]
AI Agent
I'll parse the spec, analyze the authentication scheme, then scan each endpoint with category-appropriate payloads.
validate_spec → 23 endpoints, 4 auth schemes, OpenAPI 3.1
spec_intelligence → 8-layer analysis: auth endpoints high priority
scan_spec → testing 23 endpoints with smart category selection
→ 2 endpoints vulnerable: POST /users (mass assignment), GET /files/{id} (IDOR via path traversal)
Two issues. POST /users accepts unexpected fields (role, is_admin) that bypass your input validation. GET /files/{id} allows path traversal in the ID parameter. Both bypassed the WAF because they're valid-looking requests at the HTTP level. I recommend adding body-field restrictions and path-normalization rules.
Setup: 30 Seconds
Add the MCP server config to your AI client. That's it. Every prompt and tool becomes available immediately.
Claude Desktop
{
"mcpServers": {
"waf-tester": {
"command": "npx",
"args": ["-y", "@waftester/cli", "mcp"]
}
}
}VS Code
{
"servers": {
"waf-tester": {
"command": "npx",
"args": ["-y", "@waftester/cli", "mcp"]
}
}
}Docker (Remote)
docker run -d -p 8080:8080 \
qandil/waftester
# Connect: http://localhost:8080/mcpWhich Prompt Should I Use?
Pick the prompt that matches your goal. Or just describe what you want in plain English. The AI will figure it out.
| Goal | Prompt | What You Say |
|---|---|---|
| Full WAF audit | security_audit | "Audit the WAF on staging" |
| Find WAF bypasses | waf_bypass | "Find XSS bypasses in stealth" |
| Compliance report | full_assessment | "PCI-DSS assessment for prod" |
| Map attack surface | discovery_workflow | "What endpoints exist on target?" |
| Research evasion techniques | evasion_research | "How can I bypass this WAF with UNION SELECT?" |
| Test against known CVEs | template_scan | "Check WAF coverage for common CVEs" |
| API-first testing | spec_security_audit | "Here's our OpenAPI spec, test it" |
Or Just Talk Naturally
You don't need to name a prompt. Say "audit my WAF" and the AI selects the right workflow. Say "find bypasses" and it picks waf_bypass. The prompts exist to structure the AI's reasoning, but you interact in plain English.
Ready to Try It?
One command to install. One command to scan. Real results in seconds.