What problem does it solve?
It solves the problem of getting an AI agent reliable, deterministic, read-only Kubernetes and GitOps cluster context that it can safely reason over without accidentally mutating anything.
Core Features & Use Cases
- MCP gateway for live, read-only tool calls: Exposes cub-scout verbs through
mcp serve so an agent host can discover tools and call them for explanations, traces, scans, and diagnostics.
- Deterministic context snapshots: Uses
context-pack --format json to generate a stable JSON snapshot suitable for prompt caching, diffs, and structured LLM reasoning.
- LLM-friendly presentation format: Applies
--presentation ai to produce consistently parseable output (and optionally paired human+AI formatting).
- Read-only safety invariant for agent calls: Ensures the MCP tool catalog is non-mutating so an LLM cannot apply/patch/delete cluster state through cub-scout.
Quick Start
Configure your agent host to connect to cub-scout's MCP server, then ask it for a deterministic snapshot via context-pack and follow up with doctor/explain/trace tool calls as needed.