AgentOpsSec

AgentOpsSec is a cohesive eight-project stack that treats agent safety as a real surface area instead of an afterthought. The aim is to give engineers visibility and guardrails around AI agents without taking away their capability.

It is built for engineers running AI coding agents who need clear control over what those agents can read, write, run, and spend in their environment.

The stack covers MCP Doctor for scanning and risk-rating MCP servers, MCP Firewall for blocking risky tool calls in real time, an Agent Flight Recorder for replayable run logs, Agent Review, MCP Radar trust scores, an Agent Sandbox for isolated execution, and Agent Cost Lens for per-run spend tracking.

The current focus is keeping the project local-first and MIT-licensed: no telemetry, an opinionated default install, and tight integration between the eight pieces so the whole stack feels like one cohesive layer instead of a pile of tools.