Paras Patel

From Scripts to Standards: Building a Production-Grade MCP Server for Kubernetes

Abstract:

Kubernetes automation has traditionally evolved through scripts, runbooks, CI jobs, dashboards, and command-line workflows that rely heavily on human context and operational judgment. As AI agents begin interacting directly with infrastructure and operational systems, these implicit assumptions can create significant reliability, security, and governance risks. This talk examines how the Model Context Protocol (MCP) can provide a standardized, production-grade control surface for Kubernetes rather than simply exposing kubectl through an AI interface. The proposed approach replaces broad, imperative automation with typed, intent-based capabilities governed by least-privilege access, strong schemas, read-before-write validation, approval gates, and comprehensive auditability. A layered reference architecture combines the MCP interface, policy and authorization controls, Kubernetes adapters, observability, safety mechanisms, and evidence capture. Particular attention is given to the concept of bounded agency: defining which operations should remain read-only, which diagnostic actions can be safely delegated to AI agents, and which remediation steps require stronger policy enforcement or human approval. The session also addresses prompt injection, identity binding, blast-radius containment, rollback strategies, transport security, and post-action verification. Production readiness is evaluated across correctness, safety, resilience, observability, and agent effectiveness. Looking ahead, intent-aware governance, multi-agent collaboration, simulation before execution, and controlled autonomous platform operations could transform Kubernetes expertise into reusable, explainable, and trusted operational capabilities.

Brief Profile:

Paras Patel is a Platform Engineering Leader at Rakuten Rewards with more than 14 years of experience in DevOps, cloud infrastructure, platform engineering, and AI-driven engineering systems. He specializes in building scalable, observable, secure, and cost-efficient platforms that enable high-performing engineering organizations. Paras has led platform strategy across Kubernetes ecosystems, GitHub runner infrastructure, API gateways, Model Context Protocol implementations, and AI-powered developer platforms, consistently improving developer productivity, system reliability, and operational efficiency. In his current role, he has implemented FinOps strategies that delivered more than 30% cost reduction through intelligent resource allocation, autoscaling, and cost-attribution models. He has also driven the adoption of AI-powered engineering systems that reduced deployment times by more than 60% while maintaining 99.99% uptime across large-scale environments. His work includes building self- service developer portals, scaling CI/CD platforms, and leading cloud-transformation initiatives from traditional infrastructure to private Platform-as-a-Service environments. Paras also brings deep expertise in observability, having designed unified monitoring platforms that integrate metrics, logs, and traces to provide end-to-end system visibility and accelerate incident detection and resolution using agentic systems. He is particularly focused on applying AI to automate engineering workflows, improve code quality, strengthen operational decision-making, and enable more intelligent infrastructure management. Beyond his professional role, Paras contributes actively to the engineering community through his newsletter, The AI Stack, technical blogs, conference speaking, and open-source work. He co-manages an active Kubernetes MCP Server repository, contributing to open-source innovation and knowledge sharing. He also publishes technical and industry-focused articles on Medium covering AI, DevOps, platform engineering, and modern infrastructure practices. In addition, Paras has reviewed three AI- focused books for Packt Publishing.