Commercial verification engagements — USD 8,000–15,000, court-ready dossier · Independent audit & conformance
CCS-aligned security scanner for MCP servers and AI agents (IETF draft-correctover-ccs). Credential exposure, SSRF, command-injection risk, missing auth. 14 checks → OWASP AISVS 1.0. Zero install:
npx correctover-scan.
We scan our own code with correctover-scan every week. Latest self-audit found real vulnerabilities (SSRF, missing input validation) — and we fixed them publicly. Read our self-audit report →
Current Security Score: 76/100 (improving weekly, target: 95+/100)
Scan finds risk in config. Runtime blocks the call.
After scanning, put enforcement on the path:
npx -y agent-runtime-guard # fail-closed tool-call intercept + Ed25519 receipts→ agent-runtime-guard · COMPARE.md
- Third-party interoperability — joint assessment merged into the EMILIA protocol
- 66 signed conformance test vectors — ccs-conformance-vectors
- Published methodology — Zenodo DOI 10.5281/zenodo.21783723
- Human audit — 116-check methodology: Agent Output Audit
correctover-scan audits MCP configuration used by Claude Code, Claude Desktop, Cursor, VS Code and other agents. It flags credential leaks, SSRF exposure, missing TLS, over-broad tools and related issues — the config problems that become RCE, exfil and injection paths. Runs locally (no upload), CI-friendly, SARIF for GitHub code scanning.
v1.4.0 adds --bundle mode: signal scanning over published npm packages and minified JS (17 checks, file+line findings).
| Layer | Tool | Job |
|---|---|---|
| Config scan (this repo) | correctover-scan | Find risky MCP / agent config before runtime |
| Runtime intercept | agent-runtime-guard | Block this tool call; emit receipt |
| Egress firewall | Pipelock-class | Mediate outbound network |
| Discovery / posture | Reco / Pluto-class | Inventory agents & permissions |
| Ecosystem data | mcp-atlas-census | 229,827 tools with schema (CC0) |
Scanners do not stop a live curl|sh. Runtime guards do not replace a full config audit. Use both.
npx correctover-scan@latestAuto-detects .cursor/mcp.json, claude_desktop_config.json, .claude/mcp.json, mcp.json, .vscode/mcp.json, and more. Local only — configs are not uploaded.
Free scanner is the top of funnel. Paid human audit + runtime enforcement is the product.
| Offering | Public anchor (USD) | Notes |
|---|---|---|
| This config / bundle scanner | Free | Local, SARIF, no upload |
| Runtime intercept + Ed25519 receipts (agent-runtime-guard) | Free | Fail-closed, offline receipts |
| Standard human audit (CCS-mapped, evidence-grade) | $2,500 | Written scoping required |
| Deep audit + PoC + remediation roadmap | $8,000+ | Written scoping required |
| Private deploy / integration support | $5,000 | Written scoping required |
| Annual retainer (monitoring + emergency) | $25,000/yr | Written scoping required |
Monetization rules (same across all Correctover surfaces):
- Public numbers are anchors and lead filters only. Formal quote always follows written scoping (async email only).
- 1:1 commercial quotes follow the internal PRICE-CARD (higher-ticket tracks). No discount below card without explicit owner approval.
- Zero phone. All communication async email.
- Payment instructions appear only after signed scope / SOW.
Contact for formal quote / scoping:
wangguigui@correctover.com
Subject line suggestion: CCS audit scoping — [your MCP / agent surface]