Which integrations are available for this server?

Enables GPU diagnostics and monitoring through tools that track GPU utilization, run [nvidia](/mcp/servers/integrations/nvidia)-smi inside GPU-enabled pods for real-time process and memory details, and check for GPU hardware errors (XID) and throttling events across the cluster. Provides monitoring tools that query [Prometheus](/mcp/servers/integrations/prometheus) metrics via OpenShift route for real-time cluster observability, including resource balance analysis, pod restart detection, GPU utilization tracking, and vLLM inference server performance metrics.

OpenShift MCP Server

A Model Context Protocol (MCP) server for OpenShift diagnostics and troubleshooting.

Features

Storage Tools

Storage Analysis: get_cluster_storage_report - comprehensive report of ephemeral storage usage on nodes, including top pod consumers.
Deep Forensics: inspect_node_storage_forensics - deep analysis of disk usage on a specific node, checking for unused images and container writable layers.
PV Capacity: check_persistent_volume_capacity - monitor persistent volume usage across namespaces with configurable thresholds.

Monitoring Tools

Resource Balance: get_cluster_resource_balance - analyze CPU and memory resource distribution across nodes.
Pod Restarts: detect_pod_restarts_anomalies - identify pods with excessive restart counts within a time window.
GPU Utilization: get_gpu_utilization - track GPU usage and identify idle GPU resources.
Inspect GPU Pod: inspect_gpu_pod - run nvidia-smi inside a GPU-enabled pod to view real-time process and memory details.
Check GPU Health: check_gpu_health - check for GPU hardware errors (XID) and throttling events across the cluster.
vLLM Metrics: get_vllm_metrics - monitor vLLM inference server performance metrics (throughput, queue size, cache usage).

All monitoring tools use Prometheus metrics via OpenShift route for real-time cluster observability.

Pod Diagnostics Tools

Pod Logs: get_pod_logs - retrieve and analyze logs from a specific pod, with support for previous container logs, tail limits, and time-based filtering.
Pod Diagnostics: get_pod_diagnostics - comprehensive health check of a pod including status, conditions, container states, restart counts, and issue detection.

Installation

# Using uv (recommended) uv tool install . # Or pip pip install .

Configuration

This server relies on the oc command line tool.

Ensure oc is installed and in your PATH.
Ensure you are authenticated (oc login ...) to your target cluster before running the server.

MCP Client Configuration

Configure the MCP server in your Claude Desktop or Gemini CLI settings:

{ "mcpServers": { "openshift-tools": { "command": "uv", "args": ["run", "openshift-mcp-server"] } } }

Example Usage

Once configured, you can ask questions like:

"Give me a summary of storage usage for all nodes"
"Check GPU utilization in the cluster"
"Diagnose pod health for my-app-pod in production namespace"

Simulated Tool Output (

# Storage Usage Report (3 nodes) ### Node: master0.example.com - **Filesystem**: Used: 36.70 Gi | Capacity: 99.44 Gi | Available: 62.74 Gi - **Image FS**: Used: 34.17 Gi - **Total Pod Ephemeral Storage**: 5.19 Gi **Top Pod Consumers:** - 2.60 Gi: `openshift-marketplace/redhat-operators-gb8ff` - 974.96 Mi: `openshift-marketplace/community-operators-fq744`

Simulated Tool Output (

### GPU Utilization Report **Total GPUs Found:** 4 | Node | GPU | Utilization | Memory Used | Status | |------|-----|-------------|-------------|--------| | `host-a:9400` | 0 | **0.0%** | 0.0% | ⚠️ Idle | | `host-a:9400` | 1 | **85.2%** | 92.3% | ✅ Active |

Simulated Tool Output (

### Pod Restart Anomalies (>5 in last 1h) | Namespace | Pod | Restarts | |-----------|-----|----------| | `ns-1` | `pod-a-7b666bd598-cvrlk` | **34** | | `ns-2` | `pod-b-6dcf7d7bb8-dw8sg` | **16** | #### 📋 Recommendations 1. **Check Logs**: `oc logs <pod> -n <namespace> --previous` 2. **Check Events**: `oc get events -n <namespace>`

Development

# Run locally uv run openshift-mcp-server

Testing the server directly

When run directly, the server expects JSON-RPC messages on standard input. You can verify the registered tools by simulating a full client handshake (Initialize -> Initialized -> Tools/List):

(echo '{"jsonrpc": "2.0", "id": 1, "method": "initialize", "params": {"protocolVersion": "2024-11-05", "capabilities": {}, "clientInfo": {"name": "test-client", "version": "1.0"}}}'; sleep 0.5; echo '{"jsonrpc": "2.0", "method": "notifications/initialized"}'; sleep 0.5; echo '{"jsonrpc": "2.0", "id": 2, "method": "tools/list"}') | uv run openshift-mcp-server

Expected output (truncated for brevity):

{"jsonrpc":"2.0","id":1,"result":{...}} {"jsonrpc":"2.0","id":2,"result":{"tools":[{"name":"get_cluster_storage_report",...},{"name":"inspect_node_storage_forensics",...},...]}}