Kubernetes

Establishes secure connections to Kubernetes clusters through the Model Context Protocol (MCP) transport layer, supporting multiple authentication methods including kubeconfig files, service account tokens, and in-cluster authentication. The KubernetesManager component loads and manages kubeconfig credentials, handles context/namespace switching, and maintains API client lifecycle across multiple cluster configurations. Supports stdio, SSE, and HTTP transports for flexible client integration patterns.

Wraps kubectl CLI commands into structured MCP tools with built-in command injection prevention through argument sanitization and schema validation. Each kubectl operation (get, apply, delete, exec, logs) is exposed as a discrete MCP tool with typed parameters that are validated before shell execution. Uses parameterized command construction rather than string interpolation to prevent shell metacharacter injection attacks.

Restricts which MCP tools are available to clients through server-side configuration, allowing operators to disable specific operations (e.g., disable pod exec, disable resource deletion). Filtering is configured at server startup and applied uniformly across all clients. Provides explicit tool availability metadata to clients.

Supports multiple MCP transport mechanisms for client integration: stdio for local CLI tools and VS Code extensions, Server-Sent Events (SSE) for browser-based clients, and HTTP for REST-style integrations. Transport selection is automatic based on client connection method. Each transport handles message framing, error handling, and connection lifecycle independently.

Integrates OpenTelemetry for distributed tracing, metrics collection, and logging across all MCP operations. Exports traces to observability backends (Jaeger, Datadog, New Relic) with automatic span creation for each tool invocation. Includes metrics for operation latency, error rates, and resource utilization. Traces include full context propagation for multi-step workflows.

Provides MCP prompts that guide users through complex Kubernetes operations with step-by-step instructions and context-aware suggestions. Prompts are dynamically generated based on cluster state and can include resource recommendations, troubleshooting steps, and deployment checklists. Implements prompt templates that clients can invoke to start guided workflows.

Supports multiple deployment patterns: NPM package installation for local development, Docker container deployment for cloud environments, and Helm chart deployment for Kubernetes-native installations. Includes environment-specific configuration through environment variables, config files, and Helm values. Manages multi-cluster configurations with context switching.

Executes kubectl get operations with structured output parsing, returning Kubernetes resources as typed JSON objects with optional filtering, sorting, and field selection. Supports querying pods, deployments, services, configmaps, secrets, and other resource types with output format negotiation (JSON, YAML, wide table). Implements server-side filtering through kubectl selectors and client-side filtering through response post-processing.

Creates and modifies Kubernetes resources by accepting YAML or JSON manifests, optionally applying Helm templating before submission to the cluster. Supports kubectl apply (declarative) and kubectl create (imperative) operations with conflict resolution strategies. Implements template mode for Helm charts, allowing parameterized resource generation before kubectl apply.

Deletes Kubernetes resources with configurable cascading deletion policies (orphan, background, foreground) and optional dry-run preview. Implements non-destructive mode that prevents deletion operations entirely, and read-only mode that blocks all write operations. Supports bulk deletion through label selectors and field selectors.

Executes commands inside running pods and retrieves pod logs through kubectl exec and kubectl logs commands. Supports multi-container pod selection, interactive command execution with stdin/stdout capture, and log filtering by container, timestamps, and line count. Implements timeout controls to prevent long-running exec sessions from blocking the MCP server.

Manages Helm releases through helm install, helm upgrade, and helm uninstall commands with values file support and template rendering. Supports Helm chart repositories, local chart paths, and inline values objects. Implements release history tracking and rollback capabilities through helm rollback operations.

Discovers available Kubernetes resource types, API groups, and cluster capabilities through kubectl api-resources and kubectl api-versions commands. Provides cluster metadata including version, node count, and available storage classes. Implements resource schema introspection for understanding resource structure and required fields.

Automatically redacts sensitive data (secrets, passwords, API keys) from kubectl output and logs before returning to MCP clients. Implements pattern-based masking for common secret types (docker-registry, generic, tls) and custom regex patterns for application-specific secrets. Redaction happens at the response layer before data reaches the client.

Enforces server-wide operation modes that restrict tool availability: non-destructive mode disables deletion and modification operations, read-only mode disables all write operations. Modes are configured at server startup and applied uniformly across all clients. Provides explicit error messages when clients attempt restricted operations.