agent-guardrail
Action-level governance for AI agents. Most guardrail solutions filter what agents *say* — this controls what they *do*. When agents have tool access (shell, APIs, file system, spend), you need a policy engine at the action layer. The MCP server wraps a self-hosted or remotely hosted Agent Guardrail instance, exposing policy evaluation, agent registration, kill switches, spend cap checks, and flight recorder queries directly to AI agents and orchestrators. Real incidents this prevents: agents spending $10K+ on API calls autonomously (AutoGPT), agents attempting to access restricted systems (ChaosGPT), and AI inventing policies with financial liability (Air Canada).
agent-guardrail
Action-level governance for AI agents. Most guardrail solutions filter what agents *say* — this controls what they *do*. When agents have tool access (shell, APIs, file system, spend), you need a policy engine at the action layer. The MCP server wraps a self-hosted or remotely hosted Agent Guardrail instance, exposing policy evaluation, agent registration, kill switches, spend cap checks, and flight recorder queries directly to AI agents and orchestrators. Real incidents this prevents: agents spending $10K+ on API calls autonomously (AutoGPT), agents attempting to access restricted systems (ChaosGPT), and AI inventing policies with financial liability (Air Canada).