uv sync                            # install deps (this is only required for local dev)
just docker-start                  # start Hatchet orchestration server in Docker (localhost:8888)
just worker                        # start local worker (runs LangGraph locally, no Docker)
just dev                           # run LangGraph Studio to visualize and debug graphs

LLM client (opencode, claude, etc.)
  │
  ├── MCP Server (kb_server.py)    ◄── stdio
  │     ├── ingest_document        ──▶ event push to Worker
  │     ├── search, get_document
  │     ├── list_documents, search_documents
  │     └── delete_document
  │
  └── MCP Server (k8s_server.py)   ◄── stdio
        ├── check_pods, get_logs, describe_pod
        ├── get_events, debug_pod, run_kubectl
        ├── get_deployments, get_statefulsets, get_daemonsets
        ├── get_services, get_ingresses, get_configmaps
        ├── get_secrets, exec_in_pod
        └── run_devops_agent       ──▶ event push to Worker

                     ┌──────────────────────────────┐
                     │  Hatchet Worker               │
                     │  ├─ knowledge_ingestion       │
                     │  ├─ k8s_devops                │
                     │  └─ k8s_tool                  │
                     │  All runs visible in dashboard │
                     └──────────────────────────────┘

{
  "mcpServers": {
    "knowledge-base": {
      "command": "uv",
      "args": ["run", "python", "src/mcp/kb_server.py"]
    },
    "k8s-devops": {
      "command": "uv",
      "args": ["run", "python", "src/mcp/k8s_server.py"]
    }
  }
}