Wraps agent governance into stdio transport so Claude Desktop can enforce policies on its own tool calls. Blocks prompt injection, masks PII, applies YAML approval gates, and writes Merkle-chained audit logs for every action. Built on the same primitives that instrument LangChain, CrewAI, and OpenAI Agents, so the policy DSL is framework-agnostic. Useful if you're running multi-step workflows in Claude Desktop and need deterministic guardrails or compliance logs without rewriting prompts. The same config that blocks a CrewAI agent from leaking secrets will block Claude Desktop's bash tool from doing the same.
claude mcp add --transport stdio acacian-aegis uvx aegis