If you're running multiple MCP servers in production and need governance over what your AI can actually do, Guardian acts as an intercepting proxy between Claude and your tools. It enforces YAML policy rules to block risky file access, shell commands, or secret leaks before they reach the real servers. You get a web dashboard showing live requests, audit logs, and cost tracking. Version 4.x adds fleet-wide features like cross-agent attack chain detection, threat exchange via MTX format, agent reputation scoring, and compliance evidence generation mapped to SOC2, HIPAA, and ISO frameworks. Install it with npm, run the onboard command to wrap your existing servers, then tune policies as your AI agents trip the rules.
claude mcp add --transport stdio rudraneel93-mcp-guardian uvx mcp-guardian