MCP Prompts#

Cloud Native MCP Server ships with built-in MCP prompts that agents can fetch before they start calling tools.

These prompts are meant for:

incident triage
workload troubleshooting
service connectivity diagnosis
rollout recovery
safe remediation
observability correlation
Argo CD delivery diagnosis
LLM application investigation

Prompt Catalog#

Prompt	Purpose
`cloud_native_incident_triage`	Cross-signal incident triage
`kubernetes_workload_diagnosis`	Diagnose one Pod or workload
`kubernetes_safe_remediation`	Guide patch, restart, scale, or delete workflows
`kubernetes_service_connectivity_diagnosis`	Diagnose service-to-pod connectivity and request failures
`kubernetes_rollout_recovery`	Investigate rollout failures and recovery options
`cloud_native_observability_correlation`	Correlate alerts, metrics, logs, traces, Sentry, and Langfuse
`argocd_delivery_diagnosis`	Diagnose GitOps delivery failures
`llm_app_observability_investigation`	Investigate LLM application failures and regressions
`prometheus_metrics_diagnosis`	Diagnose targets, queries, alerts, and rules in Prometheus
`loki_log_investigation`	Diagnose logs and LogQL selectors in Loki
`jaeger_trace_investigation`	Diagnose traces and dependencies in Jaeger
`grafana_dashboard_diagnosis`	Diagnose dashboards, panels, datasources, and rendering in Grafana
`alertmanager_alert_triage`	Triage alerts, groups, silences, and receivers in Alertmanager
`helm_release_diagnosis`	Diagnose Helm release state, values, manifests, and rollback choices
`kibana_log_diagnosis`	Diagnose Kibana logs, dashboards, data views, alerts, and saved objects
`elasticsearch_cluster_diagnosis`	Diagnose Elasticsearch health, nodes, indices, and search issues
`nacos_config_service_diagnosis`	Diagnose Nacos config, service discovery, and node state
`sentry_issue_investigation`	Investigate Sentry issues and issue events
`langfuse_llm_trace_investigation`	Diagnose Langfuse traces, prompts, scores, datasets, and metrics
`opentelemetry_collector_diagnosis`	Diagnose collector config, health, and pipeline state
`utilities_helper_usage`	Use helper prompts for time, sleep, and lightweight fetch workflows
`cloud_native_question_resolution`	Interpret a user question, classify the task, and route to the right service tools
`multi_service_root_cause_analysis`	Composite RCA across multiple services and observability signals
`release_regression_diagnosis`	Composite diagnosis for release or rollout regressions
`telemetry_gap_diagnosis`	Composite diagnosis for missing telemetry across collectors and backends
`end_to_end_request_path_diagnosis`	Composite diagnosis for user-facing request failures across components

Operating Model#

The prompts follow a few fixed rules:

start read-only first
prefer summary tools before full-detail tools
use exact runtime tool names
separate facts from inference
require explicit confirmation before state-changing actions

Suggested Client Flow#

List prompts from the MCP server.
Select the prompt that best matches the user request.
Fetch the prompt with arguments such as namespace, kind, name, symptom, or time_range.
Let the agent follow the embedded workflow and tool ordering.

Availability#

Prompt availability follows enabled services:

aggregate endpoints expose the full prompt catalog, filtered by enabled services
service-specific endpoints expose only prompts that match that service
prompts whose required services are disabled are filtered out or rejected by middleware