LUMINO MCP Server is an open-source MCP (Model Context Protocol) server designed to provide SREs with intelligent observability, predictive analytics, and AI-driven automation across Kubernetes, OpenShift, and Tekton environments. It exposes 39 specialized tools through the Model Context Protocol, enabling AI assistants to monitor cluster health, resources, and pipeline status in real time; analyze logs, events, and anomalies using statistical and machine learning techniques; perform automated root cause analysis on failed pipelines; predict resource bottlenecks and potential issues; and simulate configuration changes before deployment to assess impact.
Key Capabilities
Kubernetes and OpenShift Operations: Namespace and Pod management, resource queries with flexible output formats, label-based resource search across clusters, OpenShift Operator and MachineConfigPool status, etcd log analysis.
Tekton Pipeline Intelligence: Pipeline and TaskRun monitoring across namespaces, detailed log retrieval with optional cleanup, root cause analysis of failed pipelines, cross-cluster pipeline tracing, CI/CD performance baselines.
Advanced Log Analysis: Intelligent log summarization with configurable detail levels, streaming analysis of large log volumes, multi-strategy hybrid analysis, NLP-based semantic search, anomaly detection with severity classification.
Prediction and Proactive Monitoring: z-score-based statistical anomaly detection, predictive log analysis for early warning, resource bottleneck prediction, certificate health monitoring with expiration alerts, TLS certificate issue investigation.
Event Intelligence: Multi-strategy intelligent event retrieval, progressive event analysis (overview to in-depth), advanced analysis with ML pattern detection, log-event correlation.
Simulation and What-If Analysis: Monte Carlo simulation of configuration changes, pre-deployment impact analysis, risk assessment with configurable tolerance, affected component identification.
Usage
Quick Start (within 2 minutes): Claude Code CLI users can directly ask Claude Code to configure the Lumino MCP server; other MCP clients can choose MCPM (recommended, command: mcpm install @spre-sre/lumino-mcp-server) or manual setup. After installation, it can be verified with a simple query.
Runtime Requirements: Python 3.10 or higher(Core runtime); MCP client (one of Claude Desktop, Claude Code CLI, Gemini CLI, Cursor IDE); Kubernetes/OpenShift access (valid kubeconfig with read permissions); RBAC permissions (to list resources such as Pods and namespaces). Optional recommendations: uv (faster dependency management than pip), MCPM (easiest installation), Prometheus (for advanced metrics and predictive capabilities).
Run modes: By default it runs in local mode (stdio transport), suitable for direct integration with MCP clients; when running inside Kubernetes, set the namespace environment variable to enable HTTP streaming, and the server automatically detects the environment and switches transport mode.
Kubernetes authentication: Automatically detects in-cluster configuration (when running inside a Pod) or local kubeconfig (using ~/.kube/config when running locally).
ML model persistence: The predictive_log_analyzer tool persists trained ML models and training data locally in ~/.lumino/, supporting reuse of models across server restarts and incremental learning based on historical failure patterns. The model ID format is predictive_log_v1_YYYYMMDD_HHMMSS. The training data SQLite database (training_data.db) contains four tables, all defining a cluster_id column to support multiple clusters.
KubeArchive integration: KubeArchive stores Kubernetes resources outside the cluster and provides a REST API for historical resource states and logs. When Pods, PipelineRuns, or TaskRuns have been garbage collected from the active cluster, LUMINO uses KubeArchive as a fallback, and the query_kubearchive tool transparently queries this archive. The endpoint is automatically discovered through a 5-step chain (KUBEARCHIVE_HOST environment variable, OpenShift Route, Kubernetes Ingress, Kubernetes Service, kubeconfig-based Route inference). During local development, if an in-cluster Service endpoint is discovered, it automatically sets kubectl port-forward (trying ports 8081–8090, then falling back to a system-assigned port), the process starts automatically and is automatically cleaned up when the server exits.
Prometheus/Thanos integration: LUMINO automatically discovers Prometheus or Thanos Query endpoints for the prometheus_query, resource_bottleneck_forecaster, and ci_cd_performance_baselining_tool tools. Thanos Query implements the Prometheus HTTP API and is preferred when available because it provides a unified deduplicated view across replicas. The auto-discovery order depends on the runtime environment: in-cluster it is Thanos service → Prometheus service → Prometheus Operator CRD → OpenShift Routes; locally/outside the cluster it is OpenShift Routes → Thanos service → Prometheus Operator CRD → Prometheus service. Authentication uses 5 methods in priority order. The endpoint cache TTL is 5 minutes.
Tool categories: Kubernetes core (5 tools), Tekton pipelines (6), log analysis (8), event analysis (3), failure analysis and RCA (2), resource monitoring (4), namespace investigation (2), certificates and security (2), OpenShift-specific (3), CI/CD performance (2), topology and prediction (3), simulation (1).
Use cases
Suitable for SRE and DevOps teams performing intelligent observability monitoring of Kubernetes clusters, Tekton pipeline debugging, log and event analysis, root cause analysis, predictive monitoring, configuration change simulation, and CI/CD performance optimization.
Limitations
Requires Python 3.10+ and an MCP client; Kubernetes features require a valid kubeconfig and RBAC read permissions; Prometheus/Thanos and KubeArchive are optional integrations. ML models accumulate disk usage over time; by default cleanup retains models for up to 30 days (minimum retention of 3 ), training data is retained for 90 days, and the manage_prediction_training_data tool must be run periodically to clean up and reclaim disk space. Other specific limitations are not yet provided.
