shinthink/blitzstrike
Blitz Strike — a universal MCP security-audit toolbelt. Reconnaissance at speed. Analysis in depth. Validation before report.
What is Blitz Strike?
Blitz Strike is a penetration-testing toolkit packaged as a server that AI coding agents (like Claude Code, Cursor, or Gemini) can call directly. It runs a three-stage security audit — recon, analysis, validation — entirely on the server side, so a single command like run_engagement can perform a full audit from any compatible AI client.
The core idea: A scan hit is just a hypothesis. Blitz Strike refuses to report a finding until it's live-tested and confirmed. This eliminates the two classic failures of automated security tools: false positives from surface-level pattern matching, and findings reported without real-world verification.
The three tiers
| Tier | Name | Phase | What it does |
|---|---|---|---|
| 1 | BLITZ | Reconnaissance | Maps the attack surface: unauthenticated entry points, dangerous code sinks, authentication boundaries |
| 2 | EAGLE-EYE | Static analysis | Traces whether a detected vulnerability is actually reachable and exploitable — not just present in code |
| 3 | STRIKE | Validation | Performs live verification (marker reflection + negative control), enforces scope, and only then reports |
How it works
The AI agent (the "brain") plans and decides; Blitz Strike provides the deterministic tools, knowledge base, and guardrails. The agent automatically classifies the target — a URL triggers the live pipeline, a filesystem path triggers the source-code pipeline — then drives recon → analyze → verify → report.
What's included
- 130 security tools cataloged (recon, exploitation, forensics, AD, mobile, etc.) with install commands and usage flags
- 317+ tool manuals and 17 engagement playbooks (web-app, API security, Active Directory, etc.)
- 58 escalation chains — step-by-step attack paths (e.g., SSRF → cloud metadata, JWT algorithm confusion) each with success criteria and negative controls
- Web3 audit — deterministic Solidity analysis covering 13 bug classes (reentrancy, tx.origin auth, signature replay, etc.) plus executable Foundry proofs
- Out-of-band proof — an interactsh listener to catch blind SSRF/XXE/SQLi (the canary must actually arrive)
- Memory system — append-only knowledge base that grows with each engagement, storing verified insights
- Intelligence layer — WAF detection (139 signatures), tech/CVE/port correlations, fuzzing payloads, nuclei templates
Quality (measured)
On a labelled benchmark corpus across PHP, JavaScript, Python, and Java: detection rate 1.0, false-positive rate 0, precision 1.0. Coverage matrix: 5 languages × sink classes = 84% coverage.
Quick start
npx -y blitzstrike doctor # verify environment
npx -y blitzstrike install # auto-register with every detected agent CLI
Then ask your agent to "audit ./src" (source) or "audit https://example.com" (live).
Tech notes
- Written in TypeScript, runs on Bun (a fast JavaScript runtime)
- Compiles to a single static binary per platform
- Zero-install distribution via
npx/bunx - MIT-licensed; most bundled datasets are MIT or Apache-2.0 licensed public sources
Credentials needed
Only FOFA_SEARCH requires credentials (FOFA_EMAIL + FOFA_KEY). Everything else works with no API keys. Optional: H1_USERNAME adds a responsible-disclosure header to outbound requests.
Related
- Project
- Project
- Project
- Project
- Dispatch