mcp-defense

MCP tool descriptions and schemas are secured using hash-pinning to prevent drift. This process validates tool manifests against a registered hash, detecting and preempting any unauthorized modifications during server adoption and runtime invocation.

Hash-drift detection is a mechanism that pins MCP tool descriptions and input schemas to a known hash. It monitors these manifests during adoption and runtime to identify and preempt dangerous configuration changes or unauthorized content alterations.

MCP tool governance enforces ADR-based approvals by generating Architecture Decision Records for new tool adoption. It applies gated registration via secrets vaults to ensure tools meet documentation and security requirements before activation.

Threat-instrumentation checks scan MCP tool manifests for instruction-like content and assess over-permission scenarios. These checks run during server adoption to identify dangerous configurations and prevent data leakage or injection risks.

Runtime safeguards apply pre-call checks and envelope-wrapped tool metadata to prevent data leakage and injection. These checks execute during runtime invocation to validate tool safety before any dangerous configuration can process data.

A secrets vault is required to enforce gated registration for MCP servers. It ensures that tool adoption is restricted through ADR documentation and hash-drift detection, preventing unsafe tools from registering without proper credential governance.