About this project

MCP Runtime Evidence (MCP-RE) is a security layer designed to protect Model Context Protocol (MCP) tool calls from forgery, replay attacks, and tampering. It implements a runtime-evidence layer using RFC 9421 HTTP Message Signatures and RFC 9530 Content-Digest to ensure the integrity and authenticity of individual MCP calls. Key capabilities include: - **Cryptographic Protection**: Implements object-level signing, freshness, and replay protection for MCP requests and responses. - **Delegated Signing**: Supports response signing via short-TTL delegated keys attested by JOSE/JWS credentials from HSM/KMS roots. - **Transport Hardening**: Utilizes Rust-native mTLS transport and provides a server-side proxy (PEP) to protect ordinary MCP servers via a sidecar architecture. - **Multi-Language Support**: Provides Python and TypeScript SDKs bound to a shared Rust core (`mcp-re-client-core`) to ensure byte-identical evidence handling across languages. - **Enterprise Integration**: Includes support for distributed replay protection via Redis, PKCS#11 key sources for HSM/KMS custody, and online certificate revocation via OCSP. - **Stateful Continuity**: Supports stateless multi-round-trip continuation, ensuring request-associated elicitation remains cryptographically bound.