weibaohui/kom

Registers multiple Kubernetes clusters into a centralized ClusterInstances registry, automatically initializing client connections, dynamic clients, API resource caches, and connection pools for each cluster. Uses a fluent builder pattern to register clusters via kubeconfig paths, in-cluster service accounts, or raw REST configs, enabling subsequent operations to target specific clusters by identifier without re-authentication or re-initialization.

Provides a fluent, method-chaining syntax for Create, Read, Update, Delete operations on Kubernetes resources (native and CRD) using a statement builder pattern. Operations are composed via chained method calls (e.g., `kom.Cluster(id).Namespace(ns).Resource(kind).List()`) that construct a query statement, then execute against the Kubernetes API via dynamic client or typed client, with support for field selectors, label selectors, and pagination.

Implements an optional caching layer for Kubernetes resource queries (list, get operations) with configurable time-to-live (TTL) per query type or globally. Cache keys are derived from query parameters (cluster, namespace, resource kind, selectors), and cached results are automatically invalidated after TTL expires or on explicit cache clear. Reduces API server load for repeated queries without sacrificing freshness.

Implements an MCP server that can operate in two transport modes: Server-Sent Events (SSE) for HTTP-based clients and stdio for process-based clients (Claude, local tools). Server handles protocol negotiation, request routing, and response serialization transparently, enabling the same Kom tools to be accessed via different transport mechanisms without code duplication.

Translates SQL-like SELECT statements into Kubernetes API queries, parsing SQL syntax (SELECT, FROM, WHERE, ORDER BY, LIMIT) and converting WHERE clauses into label selectors and field selectors that execute against the Kubernetes API. Supports filtering by resource type, namespace, labels, fields, and result ordering/pagination, enabling non-Go developers or scripts to query clusters without learning client-go or fluent API syntax.

Provides high-level controllers for common Pod operations including remote command execution (exec), log streaming, port forwarding, and file upload/download. Wraps kubectl exec/logs/port-forward functionality via client-go's remotecommand and streaming APIs, handling stream setup, error handling, and cleanup automatically without requiring users to manage raw WebSocket or SPDY connections.

Provides controllers for Deployment lifecycle operations including rolling updates, rollback, status monitoring, and replica scaling. Tracks rollout progress by polling Deployment status (replicas ready, updated, available) and ReplicaSet history, enabling programmatic wait-for-rollout patterns and automatic rollback on failure detection without manual kubectl rollout commands.

Provides controllers for Node-level operations including node cordoning/uncordoning, draining, and topology inspection (labels, taints, capacity, allocatable resources). Enables programmatic node lifecycle management for cluster maintenance, autoscaling, or infrastructure changes without kubectl drain/cordon commands, with built-in pod eviction handling and grace period management.

Extends fluent API and CRUD operations to work with Custom Resource Definitions (CRDs) transparently, using dynamic client to handle arbitrary CRD schemas without pre-generated typed clients. Automatically discovers CRD API groups and versions via Kubernetes discovery API, enabling list, get, create, update, delete operations on any CRD without code generation or schema knowledge.

Applies or deletes Kubernetes resources from YAML manifests using server-side apply semantics (kubectl apply equivalent), parsing YAML into unstructured objects and submitting them to the Kubernetes API with field ownership tracking. Supports multi-document YAML files, variable substitution, and cascading deletes with configurable deletion policies.

Provides a callback system that executes custom hooks before and after resource operations (CRUD, pod operations, deployments), enabling extension points for logging, validation, audit trails, or custom side effects. Callbacks are registered per operation type and receive operation context (cluster, resource, action) and can modify behavior or trigger external integrations.

Exposes Kom functionality as an MCP (Model Context Protocol) server that implements ~50 built-in tools covering common DevOps operations (list pods, get logs, exec, deploy, scale, etc.). Server supports both Server-Sent Events (SSE) and stdio transports, enabling integration with AI agents, Claude, or other MCP clients without requiring direct Go SDK usage.