KT Cloud Aims to Cut Deployment Failure Root Cause Analysis Time by 90% Using MCP

The KT Cloud Platform Engineering team revealed a structure that connects AI with major DevOps tools such as Kubernetes, GitHub, ArgoCD, and Slack using MCP (Model Context Protocol). Previously, due to infrastructure isolation, engineers had to manually collect information by moving between multiple systems, but MCP provides a channel for AI to directly query external systems.

Key Use Cases and Effects

The engineering team selected 10 core MCP servers based on work frequency and impact. In particular, using Kubernetes MCP allows for correlated analysis of abnormal Pod states, events, and logs with a single query, reducing the time to derive failure cause hypotheses from the previous 15 minutes to just seconds. Additionally, AWS MCP and Terraform MCP enable the identification of idle resources, analysis of cost surge causes, and pre-deployment Plan analysis and risk assessment.

Principles for Applying to the KT Cloud PLATFORM Environment

To comply with internal network separation and security regulations, the following principles are applied:

• Read-Only Access: Data is queried with least-privilege tokens via internal security gateways (such as Tag Agent).
• Audit Logging: All tool call arguments, response data sizes, and timestamps are fully recorded.
• Data Masking: Before MCP servers send data to the LLM, PII and secret information are masked at the code level.

Troubleshooting Scenario

A scenario is presented where Slack notifications, GitHub Actions logs, ArgoCD status, and Kubernetes resource status are cross-verified through a single interface upon deployment failure. This eliminates multi-stage context switching, reducing troubleshooting time to the one-minute range and targeting an overall 90% time savings.