Your first scan

1. Open app.pencheff.com and sign in.
2. Register target → enter the URL, optional credentials, scope, exclude paths.
3. Pick a profile (standard is the default balanced OWASP Top 10 run).
4. Click Start scan — you’ll be redirected to a live progress page with SSE-streamed stages.
5. When it finishes, browse findings by severity, OWASP category, compliance framework.

> Pentest https://juice-shop.herokuapp.com — standard profile, report to slack

I'll run a full standard scan now.

An MCP-driven host (Cursor, Continue, Cline, Zed, …) uses the MCP tools in this order (fully automatic):

1. pentest_init → new session
2. check_dependencies → see which external tools are available
3. recon_passive → recon_active → recon_api_discovery
4. scan_waf → payload_generate
5. scan_injection → test_endpoint (manual exploitation)
6. scan_client_side → scan_auth → scan_authz → scan_advanced
7. exploit_chain_suggest → test_chain
8. generate_report and export_to_slack

The agent will pause to explain each finding and prove exploitability with a PoC before reporting it.

LLM endpoints are a third target kind alongside URL and Repo. Register the chat-completions URL once, paste your API key as an Authorization: Bearer … header, pick a red-team profile, and Pencheff fires all 10 OWASP LLM Top 10 (2025) modules at it.

pencheff llm-redteam \
  --target https://your-llm.example.com/v1/chat/completions \
  --provider openai-chat \
  --model your-model-id \
  --header "Authorization=Bearer your-token" \
  --profile standard \
  --strategies base64,jailbreak,crescendo,leetspeak \
  --datasets donotanswer,harmbench \
  --guardrails pii,secrets,unsafe-code,tool-authz \
  --judge-provider openai-moderation \
  --judge-endpoint https://your-judge.example.com/moderations \
  --max-rps 0.3 \
  --max-cost-usd 5 \
  --output-format html \
  --output-file llm-report.html \
  --fail-on high

In the SaaS UI:

1. Register target → pick LLM endpoint.
2. Endpoint URL = the chat-completions URL (not the model info page — a common mistake).
3. Add an Authorization header row with the value Bearer sk-… (the literal word Bearer plus a space, then your key).
4. Pick provider, model, and a profile (quick / standard / deep).
5. Optionally configure a judge (openai-moderation is recommended for reasoning-model targets), an attacker LLM for PAIR, an embedder for semantic verdicts, and rate limits (default Max RPM = 18 to stay under OpenRouter’s free-tier cap).

Findings are mapped to OWASP LLM Top 10, MITRE ATLAS, NIST AI RMF, and EU AI Act. Compare two scans at /scans/compare?a=…&b=… to gate PRs on safety regressions.