MCP Core Defense — Open-source 7-phase security proxy for MCP

I've built MCP Core Defense, a defense-in-depth security proxy interposed between AI agents and MCP servers. It implements 7 sequential verification phases:

1. Policy Engine — Deny-by-default allowlist with wildcards
2. Schema Validator — Strict JSON schema validation
3. DCI Checker — Description-Consistency via AST analysis (Python + JS/TS)
4. TDP Detector — Tool Description Poisoning scan (exfil/exec/obfuscation patterns)
5. Mutual TLS — Certificate verification + SHA256 pinning + MITM detection
6. Sandbox — Filesystem jail with path traversal prevention
7. SDK Adapter — Async MCP client integration with dry-run mode

Key features

• 127 tests passing on Python 3.10/3.11/3.12
• Cortocircuitable — stops at first failure
• Research-based — based on Shi et al. 2026, Liu et al. 2026, Zhou et al. 2026

Links

• GitHub: https://github.com/amurlaniakea/mcp-core-defense
• License: AGPL-3.0

Happy to answer questions or collaborate!