Main
Treat output as a baseline: review policies before deployment and narrow resources/conditions to your actual boundaries.
Use CLI explanations to trace why actions were included; keep that explanation as evidence in code review.
Prefer MCP integration for agent workflows, but still enforce least privilege at the IaC layer (ARNs, boundaries, SCPs).
Source-backed notes
- README lists three CLI commands:
generate-policies,fix-access-denied, andmcp-server. - README notes
mcp-serversupportsstdio(default) andhttptransports (example:--transport http). - README describes deterministic local code analysis to generate identity-based policies across multiple languages.
FAQ
- Does this guarantee least privilege?: No — it generates a baseline; you still need review and resource scoping.
- What if resources are chosen at runtime?: README notes it can’t predict dynamic values; add conditions/ARN constraints manually.
- Should I use CLI or MCP?: Use CLI for one-offs; use MCP when an agent should call tools during iteration.