Fail-closed MCP/tool-call runtime interceptor with offline-verifiable Ed25519 receipts. Zero runtime deps: npx -y agent-runtime-guard
-
Updated
Nov 8, 2026 - JavaScript
Fail-closed MCP/tool-call runtime interceptor with offline-verifiable Ed25519 receipts. Zero runtime deps: npx -y agent-runtime-guard
MCP and AI agent security scanner — config + bundle checks, SARIF, OWASP AISVS. npx correctover-scan
Publicly verifiable code provenance: manifest + signed receipt + Sigstore Rekor anchor, one GitHub Action. Evidence of existence and timestamp.
Curated list of AI agent security tools, papers and datasets: runtime controls and authorization, sandboxes, scanners for MCP servers and skills, static analysis, red teaming and evaluation, incident records, threat modelling, standards (OWASP, NIST, MITRE ATLAS). Every entry verified on GitHub. CC0, as awesome lists are.
Risk Gazetteer for MCP servers - evidence-bearing money/credential exposure labels over a SHA-256-frozen census (42,163 servers / 229,827 params). Schema v0.1 (CC0). Companion to Zenodo DOI 10.5281/zenodo.24028716.
CCS Runtime Verifier — 7-dimension AI agent tool-call verification (Structure, Schema, Latency, Cost, Identity, Integrity, Security). Ed25519 signed receipts + Sigstore Rekor anchoring. PyPI package.
GitHub Action: scan MCP (Model Context Protocol) configuration files for credential leaks, SSRF and missing auth in CI — SARIF output, fails builds on critical findings.
Largest known live MCP tool census: 229827 tools with input_schema from 13120 servers (CC0)
Correctover SDK (npm: correctover) — AI agent runtime authorization & CCS 7-dimension evidence verification. Published-source mirror; proprietary LICENSE, free for integration.
Transparent stdio proxy for MCP servers — fail-closed agent runtime interception: denies, verifier errors and timeouts are never forwarded. Ed25519 receipts. Agent 运行时全栈拦截.
Side-effect-only minimal reference: JCS canonicalization, claim lifecycle, signed receipts, external anchor. Zero dependencies.
🔥🔒 Awesome MCP (Model Context Protocol) Security 🖥️
Manual security audit service for AI agents and MCP integrations — semantic intent review of tool permissions, command injection, SSRF and credential-exposure paths, with a free 1-page audit summary.
To associate your repository with the correctover topic, visit your repo's landing page and select "manage topics."