OpenShift 4 MCP Server
Comprehensive MCP server for OpenShift/Kubernetes operations, exposing 216 tools, 7 resources, and 10 runbook prompts to manage clusters via natural language.
Cluster management: get cluster info, version, operator status, nodes (cordon/drain), namespaces, etcd health, events.
Workloads: pods (logs, exec, describe, delete), deployments (scale, rollout restart/undo/status), statefulsets, daemonsets, jobs, cronjobs, deploymentconfigs.
Networking: services, routes (create/delete), ingresses, network policies, ingress controllers, cluster network config.
Storage: persistent volumes and claims (CRUD), storage classes, volume snapshots.
Config: ConfigMaps (CRUD), Secret key listing (values never exposed), ServiceAccounts.
RBAC: users, groups, roles, role bindings, access checks, grant/revoke roles.
Builds: BuildConfigs, builds, image streams, start builds, logs.
Operators/OLM: subscriptions, catalog sources, install plans (approve), operator conditions.
Machines: MachineSets (scale), MachineConfigs, MachineConfigPools.
Monitoring: PromQL queries, Alertmanager alerts/silences, PrometheusRules.
Security: SCCs (list/create/assign), OAuth config, pod security violations.
Autoscaling: HPA, VPA, ClusterAutoscaler, MachineAutoscaler.
GitOps: ArgoCD Applications sync/health/refresh, AppProjects.
Pipelines: Tekton Pipelines/PipelineRuns, tasks, triggers.
Service Mesh: SMCP, VirtualServices, DestinationRules, PeerAuthentications, Gateways.
OpenShift AI: DSCI/DSC, notebooks, KServe inference services, model registries.
Virtualization: VMs (start/stop/restart/migrate/snapshot/create/delete).
Konflux: applications, components, snapshots, integration tests, releases.
ACM: managed clusters, policies, placements, manifest works.
Generic: apply manifests, run oc commands, list CRDs, delete resources.
Resources: live cluster info, nodes, operators, alerts, pods, events, deployments via
ocp://URIs.Prompts: SRE runbooks for pod troubleshooting, network debugging, cluster upgrades, ML deployment, etc.
Integrates with Red Hat Advanced Cluster Management to manage managed clusters, placements, policies, cluster claims, and manifest work deployments.
Enables GitOps management of ArgoCD Applications, including sync, health checks, refresh, and prune operations.
Provides access to etcd cluster status and health information.
Provides comprehensive management for Kubernetes clusters, including workloads, networking, storage, config, RBAC, autoscaling, and generic resource operations.
Provides monitoring capabilities through PromQL queries, alert management, silences, and PrometheusRules.
Manages Red Hat OpenShift clusters, including OpenShift-specific resources such as Routes, BuildConfigs, SCCs, Operators, MachineSets, and cluster version/upgrade operations.
Allows managing Tekton Pipelines, PipelineRuns, Tasks, TaskRuns, and EventListeners, including starting and canceling runs.
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@OpenShift 4 MCP Server Why is my pod crashlooping in namespace prod?"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
OpenShift & Kubernetes MCP Server
A comprehensive Model Context Protocol (MCP) server that exposes 216 tools, 7 resources, and 10 runbook prompts for cluster operations — all driven by an LLM. Works with OpenShift 4 and vanilla Kubernetes; OpenShift-specific tools (Routes, BuildConfigs, SCCs, OLM, Machines, RHOAI, Virtualization) return a clear error on plain Kubernetes clusters that don't have those APIs.
Connect it to Claude (Desktop, Code, or API) and ask natural-language questions like:
"Why is my pod crashlooping in namespace prod?" "Scale the frontend deployment to 5 replicas." "Show me all firing alerts and create a 4-hour silence for the watchdog." "Live-migrate VM database-0 to another node." "Deploy llama-3 with KServe in the ds-team namespace." "What's the status of my Tekton pipeline run in namespace ci?" "Show me all Konflux components and their latest snapshot status."
Table of Contents
Related MCP server: OpenShift SRE Copilot
Features
Domain | Tools | What you can do |
Cluster | 17 | ClusterVersion, upgrade status, nodes, cordon/drain, namespaces, etcd health, events |
Workloads | 19 | Pods (logs, exec, describe), Deployments (scale, rollout, undo), StatefulSets, DaemonSets, Jobs, CronJobs, DeploymentConfigs |
Networking | 12 | Services, OpenShift Routes (TLS), Ingress, NetworkPolicies, IngressControllers |
Storage | 10 | PVs, PVCs (create/delete), StorageClasses, VolumeSnapshots |
Config | 8 | ConfigMaps, Secrets (keys only — values never exposed), ServiceAccounts |
RBAC | 13 | OCP Users/Groups, Roles, ClusterRoles, RoleBindings, |
Builds | 8 | BuildConfigs, start/log builds, ImageStreams and tags |
Operators (OLM) | 10 | CSVs, Subscriptions, CatalogSources, InstallPlans (approve), OperatorConditions |
Machines | 11 | MachineSets (scale), Machines, MachineConfigs, MachineConfigPools (pause/unpause) |
Monitoring | 10 | PromQL instant/range queries, Alertmanager alerts/silences (CRUD), PrometheusRules |
Security | 9 | SCCs (list/create/assign), OAuth config, pod security violations |
Autoscaling | 8 | HPA (create/delete), VPA recommendations, ClusterAutoscaler, MachineAutoscaler |
GitOps | 7 | ArgoCD Applications (sync, health, refresh), AppProjects, registered clusters |
Pipelines | 10 | Tekton Pipelines/PipelineRuns/Tasks/TaskRuns, start/cancel, EventListeners |
Service Mesh | 8 | SMCP status, VirtualServices, DestinationRules, PeerAuthentications, Gateways |
OpenShift AI | 13 | DSCI/DSC status, Notebooks (start/stop), KServe InferenceServices, DSP, ModelRegistry |
Virtualization | 15 | VMs (start/stop/restart/pause/create/delete), live migration, DataVolumes, snapshots |
Konflux | 11 | Applications, Components, Snapshots, IntegrationTestScenarios, ReleasePlans |
ACM | 11 | ManagedClusters, Policies, Placements, ManifestWorks (deploy to managed clusters) |
Generic | 6 |
|
MCP Resources | 7 | Live cluster URIs: |
MCP Prompts | 10 | SRE runbooks: troubleshoot pod, upgrade cluster, debug network, deploy ML model, and more |
Requirements
Python 3.11+
ocCLI in PATH (for operations that use it; many tools fall back to direct k8s API calls)virtctlin PATH (for VM pause/unpause; optional)Access to an OpenShift 4.x cluster
Installation
git clone https://github.com/ay-garg/openshift-mcp-server.git
cd openshift-mcp-server
python3 -m venv .venv
.venv/bin/pip install -e .Authentication
The server supports five auth modes, tried in priority order:
1. OCP_CLUSTERS — multi-cluster JSON (highest priority)
See the Multi-cluster section below.
2. Service Account Token (recommended for CI/CD)
export OCP_API_URL=https://api.mycluster.example.com:6443
export OCP_TOKEN=sha256~xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxGet a long-lived token:
oc create serviceaccount mcp-server -n default
oc adm policy add-cluster-role-to-user cluster-admin -z mcp-server -n default
oc create token mcp-server -n default --duration=8760h3. Username / Password
export OCP_API_URL=https://api.mycluster.example.com:6443
export OCP_USERNAME=kubeadmin
export OCP_PASSWORD=xxxx-xxxx-xxxx-xxxxThe server runs oc login and extracts the resulting bearer token automatically.
4. kubeconfig (default for local dev)
# Uses ~/.kube/config automatically, or:
export OCP_KUBECONFIG=/path/to/kubeconfig
export OCP_KUBECONFIG_CONTEXT=my-cluster-admin # optional context name5. In-cluster (when running inside a pod)
No env vars needed — uses the mounted ServiceAccount token automatically.
TLS verification
# Disable TLS verification for the Kubernetes API connection (k8s client):
export OCP_SKIP_TLS_VERIFY=true # for self-signed certs in dev/lab clusters
# Disable TLS verification for Prometheus/Alertmanager HTTP calls:
export OCP_VERIFY_SSL=falseThese are two independent settings — OCP_SKIP_TLS_VERIFY controls the kubernetes Python client (API calls), OCP_VERIFY_SSL controls HTTP requests to Prometheus and Alertmanager.
Multi-cluster
Set OCP_CLUSTERS to a JSON array of named cluster configs:
export OCP_CLUSTERS='[
{"name": "prod", "api_url": "https://api.prod.example.com:6443", "token": "sha256~prod..."},
{"name": "staging", "api_url": "https://api.staging.example.com:6443", "token": "sha256~staging..."},
{"name": "lab", "api_url": "https://api.lab.example.com:6443", "token": "sha256~lab...", "skip_tls_verify": true}
]'Each cluster config object supports:
Field | Required | Description |
| yes | Logical name used in the |
| yes | API server URL ( |
| one of token/user+pass | Bearer token |
| one of token/user+pass | Credentials for |
| no | Set |
Then pass cluster="prod" to any tool:
list_pods(namespace="kube-system", cluster="prod")
scale_deployment(name="api", replicas=3, namespace="default", cluster="staging")Monitoring / Prometheus
By default the server auto-derives the Alertmanager URL from OCP_PROMETHEUS_URL. Override if needed:
export OCP_PROMETHEUS_URL=https://thanos-querier.openshift-monitoring.svc:9091
export OCP_ALERTMANAGER_URL=https://alertmanager-main.openshift-monitoring.svc:9093
export OCP_PROMETHEUS_TOKEN=sha256~... # defaults to OCP_TOKENEnvironment variables reference
Variable | Default | Purpose |
| — | API server URL for single-cluster token/password auth |
| — | Bearer token for the service account or user |
| — | Username for |
| — | Password for |
|
| Set |
|
| Path to a kubeconfig file |
| — | Named context within the kubeconfig |
| — | JSON array of multi-cluster configs (see above) |
| auto-detected | Prometheus/Thanos querier URL |
| auto-derived | Alertmanager URL |
|
| Token for Prometheus/Alertmanager HTTP calls |
|
| Set |
|
|
|
|
| Bind address for streamable-http transport |
|
| Port for streamable-http transport |
|
| Bind address for the Gradio web UI |
|
| Port for the Gradio web UI |
|
| Set |
| — | Required for the AI Chat tab in the Gradio UI |
|
| Model for the AI Chat tab |
Usage with Claude
Claude Code (this repository)
The .claude/settings.json already wires the server up. Open this directory in Claude Code and the ocp MCP server is available automatically.
Claude Desktop
Add to ~/Library/Application Support/Claude/claude_desktop_config.json (macOS):
{
"mcpServers": {
"ocp": {
"command": "/path/to/ocp-mcp-server/.venv/bin/python",
"args": ["-m", "ocp_mcp.server"],
"env": {
"PYTHONPATH": "/path/to/ocp-mcp-server/src",
"OCP_API_URL": "https://api.mycluster.example.com:6443",
"OCP_TOKEN": "sha256~..."
}
}
}
}Streamable HTTP transport (for remote use or web apps)
Note:
MCP_HOSTdefaults to127.0.0.1(loopback-only, with DNS-rebinding protection enabled by the MCP SDK). SetMCP_HOST=0.0.0.0explicitly when you need external access.
export MCP_TRANSPORT=streamable-http
export MCP_HOST=0.0.0.0 # bind to all interfaces for remote access
export MCP_PORT=8080
.venv/bin/ocp-mcp-serverThen point your MCP client at http://your-host:8080/mcp.
Web UI (Gradio)
A browser-based UI with two tabs — no MCP client required.
Install UI dependencies:
.venv/bin/pip install -e ".[ui]"Run:
export OCP_API_URL=https://api.mycluster.example.com:6443
export OCP_TOKEN=sha256~...
.venv/bin/ocp-mcp-ui
# Opens at http://localhost:7860Tab 1 — Tool Playground: Select any of the 216 tools from a searchable dropdown, fill in parameters, and run it directly against your cluster. Results appear instantly — no AI in the loop.
Tab 2 — AI Chat: Natural-language chat backed by Claude. Set ANTHROPIC_API_KEY and ask anything — Claude will automatically call the right OCP tools and show you what it did.
Interactive Chat Client (mcp_chat.py)
mcp_chat.py is a standalone terminal chat client that connects to any running MCP server and drives an agentic loop using the LLM of your choice. All configuration is prompted at startup — no environment variables required, though they are used as defaults when present.
Supported LLM providers
Provider | Auth |
Anthropic API | API key |
Google Vertex AI | GCP project ID + region (GCP ADC — |
Ollama | Base URL (local or remote) |
OpenAI-compatible | Base URL + optional API key (OpenAI, LM Studio, vLLM, llama.cpp, …) |
Install
pip install mcp anthropic "anthropic[vertex]" openai httpxRun
python mcp_chat.pyThe script walks you through setup interactively:
╔══════════════════════════════════════════════════════════════╗
║ MCP Chat — Setup ║
╚══════════════════════════════════════════════════════════════╝
MCP server URL [http://localhost:8080/mcp]:
LLM provider
1. Anthropic API (API key)
2. Google Vertex AI (GCP project ID + region, GCP ADC auth)
3. Ollama (local or remote)
4. OpenAI-compatible (OpenAI / LM Studio / vLLM / llama.cpp / …)
Choice:Self-signed / internal CA certificates
When an HTTPS URL is entered (for the MCP server or the model endpoint) the script asks whether the certificate is CA-signed or self-signed:
The MCP server URL is using HTTPS.
Does it use a valid CA-signed certificate? (answer 'n' for self-signed / internal CA) [Y/n]:Answering n disables SSL verification for that endpoint automatically. This is the correct answer for:
CRC (CodeReady Containers) — uses a self-signed router CA
Self-hosted OpenShift clusters with internal PKI
Local Ollama or OpenAI-compatible servers fronted by nginx with a self-signed cert
Note: For the MCP server connection, SSL verification is disabled by patching
httpx.AsyncClientfor the duration of the session (the MCP SDK does not expose averify=parameter directly). For Ollama/OpenAI-compatible clients,httpx.Client(verify=False)is passed directly. Anthropic API and Google Vertex AI always use CA-signed certificates and are never prompted.
Environment variable defaults
All prompts use environment variables as pre-filled defaults so repeat runs need fewer keystrokes:
Prompt | Env var |
MCP server URL |
|
Anthropic API key |
|
Model (Anthropic / Vertex) |
|
GCP project ID |
|
GCP region |
|
Ollama base URL |
|
Ollama model |
|
OpenAI base URL |
|
OpenAI API key |
|
OpenAI model |
|
Example session (Vertex AI + CRC cluster)
# Port-forward the deployed MCP server
oc port-forward svc/ocp-mcp-server 8080:8080 -n ocp-mcp &
python mcp_chat.py
# MCP server URL [http://localhost:8080/mcp]: https://ocp-mcp-server-ocp-mcp.apps-crc.testing/mcp
# The MCP server URL is using HTTPS.
# Does it use a valid CA-signed certificate? [Y/n]: n
# ⚠ SSL verification disabled for MCP server (self-signed cert).
# LLM provider → 2 (Google Vertex AI)
# GCP project ID: my-gcp-project
# Region [us-east5]:
# Model [claude-opus-4-8]:
# Ready — 216 tools available | provider: vertex | model: claude-opus-4-8
You: What nodes are in my cluster and are any under memory pressure?
→ list_nodes({})
→ get_node_conditions({"node":"crc-xxxxx-master-0"})Container & OpenShift Deployment
This section covers building the container image and deploying to OpenShift or any Kubernetes cluster.
Prerequisites
Podman or Docker for building/pushing the image
Access to a container registry (Quay.io, OpenShift internal registry, etc.)
ocCLI logged in to your cluster
1. Build the image
# Clone and enter the repo
git clone https://github.com/ay-garg/openshift-mcp-server.git
cd openshift-mcp-server
# Build with Podman (recommended for OpenShift)
podman build -f Containerfile -t quay.io/your-org/ocp-mcp-server:latest .
# Multi-arch build (amd64 + arm64)
podman buildx build \
--platform linux/amd64,linux/arm64 \
-f Containerfile \
-t quay.io/your-org/ocp-mcp-server:latest .
podman push quay.io/your-org/ocp-mcp-server:latestBuild arguments:
Argument | Default | Description |
|
| OpenShift CLI version; e.g. |
|
| KubeVirt virtctl version |
|
| CPU architecture: |
# Pin specific CLI versions
podman build -f Containerfile \
--build-arg OC_VERSION=4.16.3 \
--build-arg VIRTCTL_VERSION=v1.4.0 \
-t quay.io/your-org/ocp-mcp-server:4.16.3 .2. Push the image
podman push quay.io/your-org/ocp-mcp-server:latestFor the OpenShift internal registry:
# Log in to the internal registry
oc registry login
IMAGE="$(oc registry info)/ocp-mcp/ocp-mcp-server:latest"
podman build -f Containerfile -t "$IMAGE" .
podman push "$IMAGE"3. Deploy to OpenShift
3a. Create the namespace
oc new-project ocp-mcp
# or:
oc apply -f deploy/namespace.yaml3b. Create the credentials Secret
The Secret holds cluster auth and the optional Anthropic API key. Never commit real values.
In-cluster deployment (server manages the same cluster it runs in — no credentials needed):
# Only set ANTHROPIC_API_KEY if you want the Gradio AI Chat tab
oc create secret generic ocp-mcp-server-credentials \
--from-literal=ANTHROPIC_API_KEY=sk-ant-xxxxxxxx \
-n ocp-mcp
# If no Anthropic key either, create an empty secret:
oc create secret generic ocp-mcp-server-credentials -n ocp-mcpExternal cluster (server is deployed elsewhere and manages a remote cluster):
# Single cluster — token auth (recommended)
oc create secret generic ocp-mcp-server-credentials \
--from-literal=OCP_API_URL=https://api.cluster.example.com:6443 \
--from-literal=OCP_TOKEN=sha256~xxxxxxxxxxxxxxxxxxxxxxxx \
--from-literal=ANTHROPIC_API_KEY=sk-ant-xxxxxxxx \
-n ocp-mcp
# Multi-cluster
oc create secret generic ocp-mcp-server-credentials \
--from-literal=OCP_CLUSTERS='[
{"name":"prod", "api_url":"https://api.prod.example.com:6443", "token":"sha256~prod..."},
{"name":"staging", "api_url":"https://api.staging.example.com:6443", "token":"sha256~staging..."}
]' \
--from-literal=ANTHROPIC_API_KEY=sk-ant-xxxxxxxx \
-n ocp-mcpTip: Generate a long-lived ServiceAccount token for the MCP server:
oc create serviceaccount mcp-server -n default oc adm policy add-cluster-role-to-user cluster-admin -z mcp-server -n default oc create token mcp-server -n default --duration=8760h
3c. Edit the image reference
Open deploy/deployment.yaml and replace the placeholder image:
image: quay.io/your-org/ocp-mcp-server:latest3d. Apply all resources
# Using kustomize (recommended)
oc apply -k deploy/
# Or apply individually
oc apply -f deploy/serviceaccount.yaml
oc apply -f deploy/clusterrolebinding.yaml
oc apply -f deploy/configmap.yaml
oc apply -f deploy/deployment.yaml
oc apply -f deploy/service.yaml
oc apply -f deploy/route.yaml3e. Verify the deployment
# Check pod status
oc get pods -n ocp-mcp -l app.kubernetes.io/name=ocp-mcp-server
# Check logs
oc logs -n ocp-mcp -l app.kubernetes.io/name=ocp-mcp-server -f
# Get the public MCP URL
oc get route ocp-mcp-server -n ocp-mcp -o jsonpath='{.spec.host}'The server is ready when you see a line like:
INFO: Started server process
INFO: Uvicorn running on http://0.0.0.0:80803b. Deploy to vanilla Kubernetes
The same manifests work on any Kubernetes cluster. The differences from the OpenShift steps above:
Use
kubectlinstead ofocUse
deploy/ingress.yamlinstead ofdeploy/route.yaml(Ingress requires an ingress controller such as nginx-ingress)Skip
deploy/namespace.yamlif your cluster auto-creates namespaces; otherwisekubectl create namespace ocp-mcp
Create the namespace and credentials
kubectl create namespace ocp-mcp
# Token auth (replace with your cluster API URL and token)
kubectl create secret generic ocp-mcp-server-credentials \
--from-literal=OCP_API_URL=https://api.k8s.example.com:6443 \
--from-literal=OCP_TOKEN=<serviceaccount-token> \
-n ocp-mcpGenerate a long-lived ServiceAccount token:
kubectl create serviceaccount mcp-server -n default
kubectl create clusterrolebinding mcp-server-admin \
--clusterrole=cluster-admin --serviceaccount=default:mcp-server
kubectl create token mcp-server -n default --duration=8760hApply the manifests
# Apply all resources except the OpenShift Route
kubectl apply -f deploy/serviceaccount.yaml
kubectl apply -f deploy/clusterrolebinding.yaml
kubectl apply -f deploy/configmap.yaml
kubectl apply -f deploy/deployment.yaml
kubectl apply -f deploy/service.yaml
kubectl apply -f deploy/ingress.yaml # Kubernetes Ingress (not Route)Edit deploy/ingress.yaml first to set the correct hostname for your cluster.
Verify
kubectl get pods -n ocp-mcp -l app.kubernetes.io/name=ocp-mcp-server
kubectl logs -n ocp-mcp -l app.kubernetes.io/name=ocp-mcp-server -f
kubectl get ingress -n ocp-mcpKubernetes compatibility note: Core tools (workloads, networking, storage, RBAC, config, monitoring, Tekton Pipelines) work on any Kubernetes cluster. Tools for OpenShift-specific APIs (Routes, BuildConfigs, SCCs, OLM, Machines, OpenShift AI, Virtualization, Service Mesh, ACM) return a clear "API not available" message on clusters where those CRDs are absent — they do not crash the server.
4. Connect an MCP client
Once deployed, point your MCP client at the Route URL:
https://<route-host>/mcpClaude Desktop (~/Library/Application Support/Claude/claude_desktop_config.json):
{
"mcpServers": {
"ocp": {
"transport": "http",
"url": "https://<route-host>/mcp"
}
}
}Claude Code (.claude/settings.json in your project):
{
"mcpServers": {
"ocp": {
"type": "http",
"url": "https://<route-host>/mcp"
}
}
}5. MCP Inspector
The MCP Inspector is a browser-based UI for exploring MCP tools, resources, and prompts at the protocol level.
Route access is not possible with the standard inspector package. Its proxy backend binds to
127.0.0.1(loopback) by design, and the browser-side JS connects to the proxy atlocalhost:SERVER_PORT. Via a Route,localhostresolves to the user's machine — not the pod — so the proxy is never reachable.oc port-forwardis required.For remote browser-based tool exploration without port-forward, use the Gradio web UI instead (see section 7 below) — it has a Tool Playground tab covering all 216 tools and works via a standard Route.
Deploy:
oc apply -f deploy/inspector.yaml -n ocp-mcpAccess via port-forward (required):
# Forward both ports — UI (6274) and proxy backend (6277)
oc port-forward svc/mcp-inspector 6274:6274 6277:6277 -n ocp-mcpOpen http://localhost:6274 in your browser, then connect with:
Field | Value |
Transport | Streamable HTTP |
URL |
|
Use the internal ClusterIP service name — the inspector proxy (inside the pod) makes the actual connection to the MCP server, not the browser.
Remove when done:
oc delete -f deploy/inspector.yaml -n ocp-mcp6. Deploy the Gradio web UI
The Gradio UI runs as a separate Deployment using the same image with OCP_MODE=ui.
Edit deploy/deployment.yaml, add a second Deployment (or patch the existing one):
# Add to the container's env section:
- name: OCP_MODE
value: "ui"
# Change containerPort to 7860 and update the Service/Route accordingly.Or run it locally:
docker compose --profile ui up7. Environment variables reference (container)
All variables from Environment variables reference apply. Container-specific additions:
Variable | Default | Purpose |
|
|
|
|
| Always set to |
|
| Set to |
8. Production checklist
Image pushed to a private registry with image pull secret configured
Credentials Secret created with real values (not the template YAML)
OCP_SKIP_TLS_VERIFYandOCP_VERIFY_SSLset correctly for your cluster's TLS postureClusterRoleBinding scoped to the minimum permissions your use case needs (see
deploy/clusterrolebinding.yaml)Route has TLS edge termination with
insecureEdgeTerminationPolicy: RedirectMCP Inspector NOT deployed (or behind port-forward only) in production
ANTHROPIC_API_KEYrotated on the schedule required by your org's secret management policyResource
requests/limitstuned to observed usage (checkoc top pod)NetworkPolicy applied to restrict ingress to the MCP port from known LLM clients only
MCP Resources
Resources expose live cluster state as URI-addressable read-only content. MCP clients can subscribe to them and display them alongside tool results.
URI | Description |
| Cluster version, infrastructure name, API URL, platform, topology, upgrade history, and available updates |
| All nodes with role, ready status, OS image, kubelet version, and age |
| All ClusterOperators sorted degraded-first with Available/Progressing/Degraded columns |
| Currently firing Alertmanager alerts, severity-sorted, with summary |
| Pods in a namespace: phase, ready containers, restarts, IP, node, age |
| Last 50 events in a namespace sorted most-recent-first |
| Deployments in a namespace: desired/ready/available/updated replicas and health conditions |
MCP Prompts
Prompts are pre-built operational runbooks that the LLM can invoke to get step-by-step guidance. Each prompt returns a structured multi-step plan that chains together the right tools automatically.
Prompt | Parameters | Purpose |
|
| Diagnose a failing or crashlooping pod: inspect status, read logs, check events, diagnose by failure pattern (CrashLoopBackOff, OOMKilled, ImagePullBackOff, Pending) |
|
| Diagnose network connectivity between pods/services: verify selectors, check endpoints, test DNS, test TCP, inspect NetworkPolicies, check Routes |
|
| Diagnose node MemoryPressure/DiskPressure/PIDPressure: check conditions, review resource usage, surface events, cordon/drain if needed |
|
| Safe upgrade pre-flight + procedure: verify operators, nodes, etcd, alerts; pause MCPs; initiate upgrade; monitor rollout; verify completion |
|
| Provision a new OpenShift project with ResourceQuota, LimitRange, default-deny NetworkPolicy, RoleBindings, and a dedicated ServiceAccount |
|
| Deploy an ML model via OpenShift AI/RHOAI: verify RHOAI, find serving runtime, create InferenceService, monitor readiness, test endpoint, configure HPA |
| — | Systematic triage for a degraded cluster: survey operators, check nodes, verify etcd, list alerts, scan events, deep-dive degraded operators |
|
| Live-migrate a KubeVirt VM: verify running state, check RWX storage, initiate VMIM, monitor progress, verify success, troubleshoot if stuck |
|
| Diagnose a stuck operator install: inspect Subscription, InstallPlan, CSV status, approve pending plans, check pod logs, verify CatalogSource |
|
| Deploy via ArgoCD: verify GitOps operator, create AppProject, configure namespace access, create Application CR, trigger sync, verify health |
Example prompts
# Cluster health
"Give me a full health summary of the cluster"
"Which ClusterOperators are degraded and why?"
"Are there any nodes in NotReady state?"
# Workloads
"List all crashlooping pods across all namespaces"
"Scale the checkout deployment to 10 replicas in namespace shop"
"Get the last 200 log lines from pod api-xyz-abc in namespace backend"
"Roll back the frontend deployment to the previous version"
# Monitoring
"Show me all critical alerts currently firing"
"Query: rate(http_requests_total[5m]) for the last hour"
"Create a 2-hour silence for AlertName=Watchdog"
# RBAC / Security
"What permissions does user john.doe have in namespace dev?"
"Grant the edit role to group platform-team in namespace staging"
"List all SCCs and which service accounts use them"
"Create a non-privileged SCC for a workload that needs setuid binaries"
# OpenShift AI
"What's the status of the DataScienceCluster?"
"List all running notebooks in the ml-team namespace"
"Deploy a scikit-learn model from s3://models/lr-v1 using KServe"
# Virtualization
"List all VMs and their current status"
"Live-migrate VM postgres-main to node worker-3"
"Take a snapshot of VM database-0 before the upgrade"
# Konflux
"What's the build status of my component frontend in workspace team-a?"
"Show me the latest snapshot and its integration test results"
# ACM
"Which managed clusters are not compliant with the security policy?"
"Show me all placements and which clusters they selected"Repository structure
ocp-mcp-server/
├── pyproject.toml # package metadata and dependencies
├── .env.example # environment variable reference
├── Containerfile # multi-stage UBI9 container image build
├── entrypoint.sh # container entrypoint (server or Gradio UI mode)
├── mcp_chat.py # universal interactive chat client (multi-provider)
├── deploy/ # OpenShift / Kubernetes manifests
│ ├── kustomization.yaml
│ ├── namespace.yaml
│ ├── serviceaccount.yaml
│ ├── clusterrolebinding.yaml
│ ├── configmap.yaml
│ ├── secret.yaml # template only — create via oc create secret
│ ├── deployment.yaml
│ ├── service.yaml
│ ├── route.yaml
│ └── inspector.yaml # optional MCP Inspector pod (port-forward access)
├── .claude/
│ └── settings.json # Claude Code MCP configuration
└── src/
└── ocp_mcp/
├── __init__.py
├── app.py # shared FastMCP server instance + port/host config
├── server.py # entry point — imports all tool modules
├── ui.py # Gradio web UI entry point (ocp-mcp-ui)
├── client.py # multi-cluster k8s client management
├── tools/
│ ├── cluster.py # ClusterVersion, nodes, namespaces, etcd
│ ├── workloads.py # Pods, Deployments, StatefulSets, Jobs
│ ├── networking.py # Services, Routes, Ingress, NetworkPolicies
│ ├── storage.py # PVs, PVCs, StorageClasses, VolumeSnapshots
│ ├── config.py # ConfigMaps, Secrets, ServiceAccounts
│ ├── rbac.py # Users, Groups, Roles, RoleBindings
│ ├── builds.py # BuildConfigs, Builds, ImageStreams
│ ├── operators.py # OLM — CSVs, Subscriptions, InstallPlans
│ ├── machines.py # MachineSets, MachineConfigs, MCPs
│ ├── monitoring.py # Prometheus queries, Alertmanager
│ ├── security.py # SCCs, OAuth, pod security
│ ├── autoscaling.py # HPA, VPA, ClusterAutoscaler
│ ├── gitops.py # ArgoCD Applications, AppProjects
│ ├── pipelines.py # Tekton Pipelines, PipelineRuns, Tasks
│ ├── service_mesh.py # SMCP, VirtualServices, DestinationRules
│ ├── ocp_ai.py # RHOAI, Notebooks, KServe, ModelRegistry
│ ├── virtualization.py # KubeVirt VMs, live migration, snapshots
│ ├── konflux.py # Konflux Applications, Components, Releases
│ ├── acm.py # ACM ManagedClusters, Policies, ManifestWorks
│ └── generic.py # apply_manifest, run_oc_command, list_crds
├── resources/
│ └── __init__.py # MCP resource URIs (ocp://cluster/info, etc.)
└── prompts/
└── __init__.py # SRE runbook prompt templatesArchitecture
LLM (Claude)
│
│ MCP protocol (stdio or streamable-http)
▼
ocp-mcp-server
│
├── client.py ──────────────────────────────────────────┐
│ ClusterRegistry │
│ ├── ClusterClient("prod") → kubernetes Python SDK │
│ ├── ClusterClient("staging") │
│ └── ClusterClient("lab") │
│ │
├── tools/*.py → @mcp.tool() │
│ All 216 tools call get_client(cluster) ───────────────┘
│ then use: k8s typed APIs (CoreV1, AppsV1, …)
│ CustomObjectsApi for OCP/OLM/RHOAI/Virt CRDs
│ subprocess oc CLI for operations not in k8s API
│
├── resources/__init__.py → @mcp.resource("ocp://…")
│ Live cluster state as URI-addressable content
│
└── prompts/__init__.py → @mcp.prompt()
SRE runbook templates the LLM can invokeAuth flow
ClusterRegistry._load() — tried in order, first success wins:
1. OCP_CLUSTERS → JSON array → one ClusterClient per entry
2. OCP_API_URL + OCP_TOKEN → bearer-token ClusterClient
3. OCP_API_URL + OCP_USERNAME + OCP_PASSWORD → oc login → extract token
4. OCP_KUBECONFIG / OCP_KUBECONFIG_CONTEXT → load_kube_config
5. In-cluster ServiceAccount tokenDesign principles
No mock data — every tool makes real API calls or runs
oc.Safe defaults — secrets never expose values, only key names. Destructive tools have
WARNINGin their docstrings so the LLM knows to confirm before executing.Graceful degradation — tools catch
ApiExceptionand return readable errors. Missing CRDs (e.g. KubeVirt not installed) return a helpful message instead of crashing.Multi-cluster first — every tool accepts a
clusterparameter. The default cluster is whichever config loaded first.Escape hatches —
apply_manifest,run_oc_command, andlist_custom_resourceslet the LLM reach anything not covered by a typed tool.
Adding a new tool
Find the relevant module in
src/ocp_mcp/tools/or create a new one.Add a function decorated with
@mcp.tool():
from ocp_mcp.app import mcp
from ocp_mcp.client import format_error, get_client
@mcp.tool()
def my_new_tool(name: str, namespace: str = "default", cluster: str = "") -> str:
"""One-sentence description shown to the LLM."""
c = get_client(cluster)
try:
result = c.core_v1.read_namespaced_something(name, namespace)
return f"Result: {result.metadata.name}"
except Exception as e:
return format_error(e)If you created a new file, add
import ocp_mcp.tools.your_moduletoserver.py.
Conventions:
Always accept
cluster: str = ""as the last parameter before any cluster-specific args.Call
get_client(cluster)and usec.oc_args()when buildingrun_ocinvocations — never callrun_ocwithout the cluster auth args, or multi-cluster calls will silently target the wrong cluster.Return strings only — tool output is text surfaced directly to the LLM.
Catch all exceptions and return
format_error(e)rather than letting them propagate.
Tool highlights
Security tools (security.py)
Tool | Description |
| List all SCCs sorted by priority |
| Full SCC detail: volumes, capabilities, users, groups |
| Create a custom SCC with parameters: |
| Grant an SCC to a ServiceAccount via |
| Revoke an SCC from a ServiceAccount |
| Grant a ClusterRole to a user |
| Grant a ClusterRole to a group |
| Get OAuth configuration and identity providers |
| Surface FailedCreate events matching SCC/security keywords |
create_scc parameters:
Parameter | Default | Description |
| required | SCC name |
|
| Allow containers to run as fully privileged (root with all capabilities) |
|
| Allow containers to use the host network namespace |
|
| Allow containers to use the host PID namespace |
|
| Sets |
|
| Allow processes to gain more privileges than their parent (required for setuid binaries like |
|
| Named cluster to target |
Generic escape-hatch tools (generic.py)
Tool | Description |
| Apply YAML/JSON manifest via |
| Get any resource in YAML, JSON, wide, or describe format |
| Delete any resource by type and name |
| List any CRD by group/version/plural |
| Escape hatch: run any |
| List all CustomResourceDefinitions |
Tekton Pipelines (pipelines.py)
Works on any Kubernetes cluster with Tekton installed (including OpenShift Pipelines).
Tool | Description |
| List Pipelines in a namespace |
| Full Pipeline spec: tasks, params, workspaces |
| List PipelineRuns with status; filter by label selector |
| PipelineRun detail: task statuses, params, start/end time, duration |
| Trigger a new PipelineRun with optional params and workspaces |
| Cancel a running PipelineRun |
| List Tasks in a namespace |
| List TaskRuns with status |
| List TriggerTemplates (webhook-driven pipeline triggers) |
| List EventListeners and their trigger bindings |
Example prompts:
"List all pipeline runs in namespace ci and show me which ones failed"
"Get the full log context for pipeline run build-frontend-xyz"
"Start pipeline build-and-push in namespace ci with IMAGE=quay.io/org/app:latest"
"Cancel the running pipeline run deploy-staging-abc"
"What triggers are configured in the platform namespace?"Konflux / RHTAP (konflux.py)
Konflux (Red Hat Trusted Application Pipeline) tools. Requires the Konflux CRDs (appstudio.redhat.com) installed on your cluster.
Tool | Description |
| List Konflux Applications in a workspace/namespace |
| Application detail: components, environments, status |
| List Components; filter by application |
| Component detail: source repo, build pipeline, container image |
| Register a new Component from a git repository |
| List Snapshots; filter by application |
| Snapshot status including all integration test results |
| List IntegrationTestScenarios for an application |
| List ReleasePlans; filter by application |
| List Releases with status and target environment |
| List PipelineRuns for a component (build history) |
Example prompts:
"What Konflux applications exist in namespace team-a?"
"Show me the latest snapshot for application frontend and its integration test results"
"List all components in application backend-api and their source repos"
"What's the build history for component api-gateway?"
"Are there any failed releases in namespace platform?"
"Show me all integration test scenarios configured for application my-app"Dependencies
Package | Purpose |
| Model Context Protocol SDK (FastMCP + streamable-http transport) |
| Kubernetes Python client (typed APIs + dynamic client) |
| HTTP client for Prometheus/Alertmanager API calls |
| YAML parsing for |
| Timestamp parsing for |
| Alternative table formatting |
| Browser-based web UI (optional — |
| Claude AI for the Chat tab (optional — included in |
Security considerations
Secrets —
get_secret_keyslists key names only.list_secretsshows type and count. Values are never returned.Destructive ops —
delete_namespace,drain_node,delete_virtual_machine, etc. includeDESTRUCTIVEwarnings in their docstrings so the LLM knows to confirm before executing.run_oc_command— usesshlex.split(no shell=True) so shell metacharacters (|,>,;) are inert literal arguments. Blocked verbs:delete,rm,exec,replace— these have typed tools with confirmation prompts.apply_manifest— applies arbitrary YAML; the LLM should show the manifest to the user before calling this in agentic contexts. The-n namespaceflag does not restrict cluster-scoped resources.Bearer token redaction —
run_ocredacts--token <value>to--token <redacted>in all error messages, preventing credential exposure in LLM context or logs.Multi-cluster routing — all
ocCLI calls prependc.oc_args()(injects--serverand--token) so the correct cluster is always targeted when multiple clusters are configured.RBAC — create a minimal ServiceAccount with only the permissions your use case needs. The tools work with whatever RBAC the token has.
License
Apache License 2.0 — see LICENSE for details.
Contributing
Issues and PRs welcome. The tool modules are intentionally kept flat and simple — one domain per file, one @mcp.tool() per operation, no shared state between tools.
Available Tools
216 toolsadd_cluster_role_to_groupB
Grant a ClusterRole to a group via 'oc adm policy add-cluster-role-to-group'.
Inputs: role (required), group (required), cluster. Returns: result
add_cluster_role_to_userB
Grant a ClusterRole to a user via 'oc adm policy add-cluster-role-to-user'.
Inputs: role (required), username (required), cluster. Returns: result
add_scc_to_service_accountA
Grant an SCC to a ServiceAccount via 'oc adm policy add-scc-to-user'. Equivalent to: oc adm policy add-scc-to-user system:serviceaccount::
Inputs: scc_name (required), namespace (required), service_account (required), cluster. Returns: result
add_user_to_groupB
Add a user to an OpenShift Group (atomic — uses oc adm groups add-users).
Inputs: username (required), group_name (required), cluster. Returns: result
apply_manifestC
Apply a YAML or JSON manifest to the cluster via 'oc apply -f -'. WARNING: This applies resources directly. Review manifests before applying.
Inputs: yaml_content (required), cluster, namespace. Returns: result
approve_install_planC
Approve a Manual InstallPlan by patching spec.approved=True.
Inputs: name (required), cluster, namespace. Returns: result
cancel_pipeline_runC
Cancel a running Tekton PipelineRun.
Inputs: name (required), cluster, namespace. Returns: result
cordon_nodeA
Cordon a node so that no new pods are scheduled on it.
Inputs: name (required), cluster. Returns: result
create_componentC
Create a Konflux Component.
Inputs: name (required), namespace (required), source_repo_url (required), application_name (required), +3 more. Returns: result
create_configmapB
Create a ConfigMap. data is KEY=VALUE pairs separated by newlines or commas. Example: KEY1=value1\nKEY2=value2
Inputs: data (required), name (required), namespace (required), cluster. Returns: result
create_groupC
Create an empty OpenShift Group.
Inputs: name (required), cluster. Returns: result
create_hpaC
Create an HPA using 'oc autoscale'. target_ref format: 'Deployment/my-app' or 'DeploymentConfig/my-dc'. cpu_percent: target CPU utilisation percentage (default 80).
Inputs: name (required), namespace (required), target_ref (required), max_replicas (required), +3 more. Returns: result
create_inference_serviceC
Create a KServe InferenceService CR.
Inputs: name (required), namespace (required), storage_uri (required), model_format (required), +3 more. Returns: result
create_job_from_cronjobA
Create a Job immediately from a CronJob template (manual trigger).
Args: cronjob_name: Source CronJob name. job_name: Name for the newly created Job. namespace: Namespace (default: "default"). cluster: Named cluster to target (empty = default).
Inputs: job_name (required), cronjob_name (required), cluster, namespace. Returns: result
create_machine_autoscalerA
Create a MachineAutoscaler CR targeting a MachineSet in openshift-machine-api. Requires a ClusterAutoscaler to be configured for this to take effect.
Inputs: name (required), max_replicas (required), min_replicas (required), machineset_name (required), +1 more. Returns: result
create_manifest_workC
Create a ManifestWork to deploy resources to a ManagedCluster.
Inputs: name (required), cluster_name (required), manifests_yaml (required), cluster. Returns: result
create_namespaceB
Create a new namespace, optionally with labels.
Args: name: Namespace name. labels: Comma-separated KEY=VALUE pairs (e.g. "env=prod,team=platform"). cluster: Named cluster to target (empty = default).
Inputs: name (required), labels, cluster. Returns: result
create_pvcB
Create a PersistentVolumeClaim. storage_request example: 10Gi access_mode: ReadWriteOnce, ReadOnlyMany, ReadWriteMany, ReadWriteOncePod.
Inputs: name (required), namespace (required), storage_request (required), cluster, +2 more. Returns: result
create_role_bindingC
Create a RoleBinding in a namespace. subject_kind: User, Group, or ServiceAccount. cluster_role: if True, references a ClusterRole instead of a Role.
Inputs: name (required), namespace (required), role_name (required), subject_kind (required), +4 more. Returns: result
create_routeA
Create an OpenShift Route by exposing a Service.
Uses 'oc expose' to create a basic route and then patches TLS termination and path if required.
Args: name: Route name. namespace: Namespace. service: Name of the Service to expose. host: Hostname for the route (empty = let the router assign one). path: URL path prefix (default "/"). tls_termination: TLS termination type: "edge", "passthrough", "reencrypt", or "" (no TLS). port: Target port name or number (empty = first port from Service). cluster: Named cluster to target (empty = default).
Inputs: name (required), service (required), namespace (required), host, +4 more. Returns: result
create_sccB
Create a custom SecurityContextConstraint CR. run_as_any: if True, sets RunAsAny for runAsUser and fsGroup. allow_privilege_escalation: if True, allows processes to gain more privileges than their parent (required for setuid binaries).
Inputs: name (required), cluster, host_pid, privileged, +3 more. Returns: result
create_silenceA
Create an Alertmanager silence. matchers: comma-separated key=value or key=~regex pairs. Example: 'alertname=Watchdog,severity=~warning|critical' duration_hours: silence duration in hours (default 4).
Inputs: comment (required), matchers (required), cluster, duration_hours. Returns: result
create_subscriptionA
Create an OLM Subscription to install an operator from a CatalogSource. install_mode is informational only (set via OperatorGroup, not Subscription spec).
Inputs: name (required), source (required), channel (required), package (required), +4 more. Returns: result
create_virtual_machineC
Create a VirtualMachine. Provide image_url (HTTP) or pvc_name (existing PVC).
Inputs: name (required), memory (required), cpu_cores (required), namespace (required), +4 more. Returns: result
create_vm_snapshotC
Create a VirtualMachineSnapshot of a VM.
Inputs: name (required), vm_name (required), namespace (required), cluster. Returns: result
create_volume_snapshotB
Create a VolumeSnapshot from a PersistentVolumeClaim.
Inputs: name (required), pvc_name (required), namespace (required), cluster, +1 more. Returns: result
delete_configmapC
Delete a ConfigMap.
Inputs: name (required), cluster, namespace. Returns: result
delete_hpaC
Delete a HorizontalPodAutoscaler.
Inputs: name (required), cluster, namespace. Returns: result
delete_inference_serviceC
Delete a KServe InferenceService.
Inputs: name (required), namespace (required), cluster. Returns: result
delete_machineA
Delete a Machine — the Machine controller will drain and terminate the underlying node. WARNING: Destructive. The node will be cordoned, drained, and terminated.
Inputs: name (required), cluster. Returns: result
delete_namespaceA
DESTRUCTIVE: Delete a namespace and ALL resources within it.
This action is irreversible. All pods, services, persistent volume claims, and other objects in the namespace will be permanently deleted.
Args: name: Namespace to delete. cluster: Named cluster to target (empty = default).
Inputs: name (required), cluster. Returns: result
delete_network_policyB
Delete a NetworkPolicy.
Args: name: NetworkPolicy name. namespace: Namespace (default: "default"). cluster: Named cluster to target (empty = default).
Inputs: name (required), cluster, namespace. Returns: result
delete_podC
Delete a pod.
Args: name: Pod name. namespace: Namespace (default: "default"). force: Force-delete immediately (grace-period=0). cluster: Named cluster to target (empty = default).
Inputs: name (required), force, cluster, namespace. Returns: result
delete_pvB
Delete a PersistentVolume. WARNING: ensure no PVC is bound before deleting.
Inputs: name (required), cluster. Returns: result
delete_pvcB
Delete a PersistentVolumeClaim. WARNING: data loss may occur if PVC is in use.
Inputs: name (required), cluster, namespace. Returns: result
delete_resourceC
Delete any Kubernetes/OpenShift resource by type and name. WARNING: Destructive operation. Confirm the resource name before proceeding.
Inputs: name (required), resource_type (required), force, cluster, +1 more. Returns: result
delete_role_bindingC
Delete a RoleBinding.
Inputs: name (required), cluster, namespace. Returns: result
delete_routeC
Delete an OpenShift Route.
Args: name: Route name. namespace: Namespace (default: "default"). cluster: Named cluster to target (empty = default).
Inputs: name (required), cluster, namespace. Returns: result
delete_serviceC
Delete a Service.
Args: name: Service name. namespace: Namespace (default: "default"). cluster: Named cluster to target (empty = default).
Inputs: name (required), cluster, namespace. Returns: result
delete_silenceB
Delete an Alertmanager silence by its ID.
Inputs: silence_id (required), cluster. Returns: result
delete_subscriptionA
Delete an OLM Subscription. Set also_delete_csv=True to also remove the currently installed CSV.
Inputs: name (required), cluster, namespace, also_delete_csv. Returns: result
delete_virtual_machineC
Delete a VirtualMachine (stop it first if running).
Inputs: name (required), namespace (required), cluster. Returns: result
describe_podB
Run 'oc describe pod' for detailed pod information including events.
Args: name: Pod name. namespace: Namespace (default: "default"). cluster: Named cluster to target (empty = default).
Inputs: name (required), cluster, namespace. Returns: result
drain_nodeA
Drain a node by evicting all evictable pods (cordons the node first).
Args: name: Name of the node to drain. ignore_daemonsets: Ignore DaemonSet-managed pods (default True). delete_emptydir_data: Allow deletion of pods with emptyDir volumes (default False). force: Force eviction of pods not managed by a controller (default False). grace_period: Seconds for graceful termination; -1 uses pod default. cluster: Named cluster to target (empty = default).
Inputs: name (required), force, cluster, grace_period, +2 more. Returns: result
exec_in_podB
Execute a command inside a pod via oc exec. command is split with shlex — shell metacharacters (pipes, redirects) are not interpreted.
Inputs: name (required), command (required), cluster, container, +1 more. Returns: result
get_alert_detailsB
Get full labels, annotations, status, and silence info for all instances of a named alert.
Inputs: alert_name (required), cluster. Returns: result
get_buildC
Get detailed Build information: phase, reason, output image, and timing.
Inputs: name (required), cluster, namespace. Returns: result
get_build_configB
Describe a BuildConfig in detail using oc describe bc.
Inputs: name (required), cluster, namespace. Returns: result
get_build_logsA
Retrieve logs for a Build. Set follow=True to stream (returns available output).
Inputs: name (required), follow, cluster, namespace. Returns: result
get_cluster_autoscalerB
Get the ClusterAutoscaler 'default' resource limits and scale-down configuration.
Inputs: cluster. Returns: result
get_cluster_infoB
Get high-level cluster information: name, API URL, version, channel, and platform type.
Inputs: cluster. Returns: result
get_cluster_network_configA
Get the cluster-wide Network configuration from config.openshift.io/v1 networks/cluster.
Shows network type, cluster networks, service networks, and machine networks.
Inputs: cluster. Returns: result
get_cluster_operator_detailsC
Get detailed conditions, versions, and related objects for a single ClusterOperator.
Inputs: name (required), cluster. Returns: result
get_cluster_operatorsB
List all ClusterOperators with Available/Progressing/Degraded/Version columns; degraded operators are listed first.
Inputs: cluster. Returns: result
get_cluster_versionB
Show ClusterVersion history, conditions (Available/Progressing/Degraded), and available updates.
Inputs: cluster. Returns: result
get_componentB
Get detailed Konflux Component info including GitOps repo and nudges.
Inputs: name (required), namespace (required), cluster. Returns: result
get_configmapA
Get ConfigMap keys and values. Values longer than 200 characters are truncated.
Inputs: name (required), cluster, namespace. Returns: result
get_data_science_clusterC
Get DataScienceCluster status: component management states.
Inputs: cluster. Returns: result
get_dsciC
Get DSCInitialization status: phase, conditions, and component enablement.
Inputs: cluster. Returns: result
get_etcd_statusC
Get ETCD cluster status from operator.openshift.io/v1 etcds/cluster.
Inputs: cluster. Returns: result
get_gitops_applicationA
Get ArgoCD Application details: sync/health/repo/path/destination + resources table.
Inputs: name (required), cluster, namespace. Returns: result
get_gitops_application_healthA
Get per-resource health status for an ArgoCD Application.
Inputs: name (required), cluster, namespace. Returns: result
get_infrastructure_configC
Get Infrastructure and Network configuration CRDs from config.openshift.io/v1.
Inputs: cluster. Returns: result
get_ingress_controllerA
Get an OpenShift IngressController from operator.openshift.io/v1.
Args: name: IngressController name (default: "default"). cluster: Named cluster to target (empty = default).
Inputs: name, cluster. Returns: result
get_konflux_applicationC
Get Konflux Application details including conditions.
Inputs: name (required), namespace (required), cluster. Returns: result
get_machineB
Get detailed Machine info: phase, nodeRef, providerID, addresses, conditions.
Inputs: name (required), cluster. Returns: result
get_machine_config_poolC
Get MachineConfigPool detail: conditions and machine counts.
Inputs: name (required), cluster. Returns: result
get_machine_setB
Describe a MachineSet showing all spec and status details.
Inputs: name (required), cluster. Returns: result
get_managed_clusterB
Get detailed ManagedCluster info: capacity, conditions, labels.
Inputs: name (required), cluster. Returns: result
get_monitoring_configB
Read the cluster-monitoring-config ConfigMap from the openshift-monitoring namespace.
Inputs: cluster. Returns: result
get_namespace_resource_quotaC
List ResourceQuotas and LimitRanges in a namespace.
Inputs: namespace (required), cluster. Returns: result
get_nodeA
Get capacity, allocatable resources, conditions, and taints for a node.
Inputs: name (required), cluster. Returns: result
get_oauth_configC
Get the cluster OAuth configuration and identity providers from config.openshift.io/v1.
Inputs: cluster. Returns: result
get_operator_statusB
Get ClusterServiceVersion status: phase, conditions, and owned CRDs. If namespace is empty, all namespaces are searched for a CSV whose name matches.
Inputs: name (required), cluster, namespace. Returns: result
get_pipelineB
Get Tekton Pipeline details: tasks with dependencies, params, and workspaces.
Inputs: name (required), cluster, namespace. Returns: result
get_pipeline_runC
Get Tekton PipelineRun details: status, duration, and child task run references.
Inputs: name (required), cluster, namespace. Returns: result
get_placement_decisionsA
List PlacementDecisions for a Placement showing selected clusters.
Inputs: namespace (required), placement_name (required), cluster. Returns: result
get_podA
Show detailed information for a pod: phase, node, IP, QoS class, container states, and conditions.
Args: name: Pod name. namespace: Namespace (default: "default"). cluster: Named cluster to target (empty = default).
Inputs: name (required), cluster, namespace. Returns: result
get_pod_logsA
Fetch logs from a pod container.
Args: name: Pod name. namespace: Namespace (default: "default"). container: Container name (empty = first/only container). previous: Return logs from the previous container instance. tail_lines: Number of lines from the end (default 100, 0 = unlimited). since_seconds: Only return logs newer than this many seconds. cluster: Named cluster to target (empty = default).
Inputs: name (required), cluster, previous, container, +3 more. Returns: result
get_policy_complianceC
Show per-cluster compliance status for an ACM Policy.
Inputs: name (required), namespace (required), cluster. Returns: result
get_pvC
Get detailed PersistentVolume info including CSI, NFS, HostPath, or EBS source.
Inputs: name (required), cluster. Returns: result
get_pvcB
Get detailed PersistentVolumeClaim info including conditions.
Inputs: name (required), cluster, namespace. Returns: result
get_resourceB
Get any resource in YAML, JSON, or wide format. Use output='describe' for oc describe.
Inputs: name (required), resource_type (required), output, cluster, +1 more. Returns: result
get_rhoai_component_statusB
Check RHOAI/OpenShift AI overall health: pods in RHOAI namespaces, DSC/DSCI status, CRD presence.
Inputs: cluster. Returns: result
get_routeA
Get detailed information for an OpenShift Route including admitted status and TLS config.
Args: name: Route name. namespace: Namespace (default: "default"). cluster: Named cluster to target (empty = default).
Inputs: name (required), cluster, namespace. Returns: result
get_sccA
Get full SCC detail: volumes, capabilities, allowedCapabilities, users, and groups.
Inputs: name (required), cluster. Returns: result
get_secret_keysA
List the key names of a Secret. Values are never shown for security.
Inputs: name (required), cluster, namespace. Returns: result
get_serviceB
Show detailed info for a Service including selector, ports, and current Endpoints.
Args: name: Service name. namespace: Namespace (default: "default"). cluster: Named cluster to target (empty = default).
Inputs: name (required), cluster, namespace. Returns: result
get_smcp_statusA
Get ServiceMeshControlPlane status: conditions and component readiness breakdown.
Inputs: name (required), namespace (required), cluster. Returns: result
get_snapshot_statusC
Get Snapshot with integration test results and component images.
Inputs: name (required), namespace (required), cluster. Returns: result
get_userA
Get OpenShift user details: full name, identities, and group memberships.
Inputs: name (required), cluster. Returns: result
get_user_accessB
List all actions a user can perform using 'oc auth can-i --list --as='. Specify namespace to scope the check; omit for cluster-wide actions.
Inputs: username (required), cluster, namespace. Returns: result
get_virtual_machineB
Get detailed VM info: running state, resources, interfaces, disks, conditions.
Inputs: name (required), namespace (required), cluster. Returns: result
get_vpa_recommendationB
Show VPA recommendations per container: lowerBound/target/upperBound/uncappedTarget.
Inputs: name (required), cluster, namespace. Returns: result
list_alerting_rulesB
List PrometheusRules (monitoring.coreos.com/v1) with rule group count.
Inputs: cluster, namespace. Returns: result
list_alertsB
List active alerts from Alertmanager, optionally filtered by severity and/or state. severity: critical | warning | info state: active | suppressed | unprocessed
Inputs: state, cluster, severity. Returns: result
list_app_projectsB
List ArgoCD AppProjects with source repos, destinations, and cluster resource access.
Inputs: cluster, namespace. Returns: result
list_build_configsC
List BuildConfigs with strategy type, source, last version, and age.
Inputs: cluster, namespace. Returns: result
list_buildsC
List Builds with phase, duration, reason, and age.
Inputs: cluster, namespace, label_selector. Returns: result
list_catalog_sourcesC
List CatalogSources with display name, type, publisher, and connection state.
Inputs: cluster, namespace. Returns: result
list_cluster_claimsC
List ACM ClusterClaims with pool and resolved cluster.
Inputs: cluster. Returns: result
list_cluster_contextsC
Show all configured MCP clusters and the active oc kubeconfig contexts.
Inputs: cluster. Returns: result
list_cluster_role_bindingsC
List ClusterRoleBindings with role reference and bound subjects.
Inputs: cluster, label_selector. Returns: result
list_cluster_rolesB
List ClusterRoles with the number of policy rules.
Inputs: cluster, label_selector. Returns: result
list_cluster_setsC
List ACM ManagedClusterSets.
Inputs: cluster. Returns: result
list_component_pipeline_runsB
List Tekton PipelineRuns associated with Konflux components.
Inputs: namespace (required), cluster, component_name. Returns: result
list_componentsB
List Konflux Components with source repo, branch, and build status.
Inputs: namespace (required), cluster, application_name. Returns: result
list_configmapsC
List ConfigMaps with namespace, name, key count, and age.
Inputs: cluster, namespace, label_selector. Returns: result
list_crdsB
List all CustomResourceDefinitions with group, stored versions, and scope.
Inputs: cluster, label_selector. Returns: result
list_cronjobsA
List CronJobs with NAMESPACE, NAME, SCHEDULE, SUSPEND, ACTIVE, LAST RUN, and AGE.
Args: namespace: Namespace to query (empty = all namespaces). label_selector: Label selector to filter CronJobs. cluster: Named cluster to target (empty = default).
Inputs: cluster, namespace, label_selector. Returns: result
list_custom_resourcesC
List any custom resource by group/version/plural — works for any CRD.
Inputs: group (required), plural (required), version (required), cluster, +2 more. Returns: result
list_daemonsetsA
List DaemonSets with NAMESPACE, NAME, DESIRED, CURRENT, READY, UP-TO-DATE, AVAILABLE, and AGE.
Args: namespace: Namespace to query (empty = all namespaces). label_selector: Label selector to filter DaemonSets. cluster: Named cluster to target (empty = default).
Inputs: cluster, namespace, label_selector. Returns: result
list_data_science_pipelinesB
List DataSciencePipelinesApplications with readiness, storage, and API endpoint.
Inputs: cluster, namespace. Returns: result
list_data_science_projectsB
List OpenShift AI Data Science Projects (namespaces with opendatahub.io/dashboard=true label).
Inputs: cluster. Returns: result
list_data_volumesC
List DataVolumes with phase, progress, capacity, and source type.
Inputs: cluster, namespace. Returns: result
list_deployment_configsA
List OpenShift DeploymentConfigs (apps.openshift.io/v1) with NAMESPACE, NAME, REVISION, DESIRED, CURRENT, and AGE.
Args: namespace: Namespace to query (empty = all namespaces). cluster: Named cluster to target (empty = default).
Inputs: cluster, namespace. Returns: result
list_deploymentsA
List Deployments with NAMESPACE, NAME, READY, UP-TO-DATE, AVAILABLE, and AGE.
Args: namespace: Namespace to query (empty = all namespaces). label_selector: Label selector to filter deployments. cluster: Named cluster to target (empty = default).
Inputs: cluster, namespace, label_selector. Returns: result
list_destination_rulesC
List Istio DestinationRules with host, subsets, TLS mode, and age.
Inputs: cluster, namespace. Returns: result
list_event_listenersB
List Tekton EventListeners with readiness, URL, and age.
Inputs: cluster, namespace. Returns: result
list_eventsA
List the most recent 50 events, optionally filtered by namespace, field selector, or involved object name.
Args: namespace: Namespace to query (empty = all namespaces). field_selector: Kubernetes field selector (e.g. "reason=BackOff"). involved_object: Filter by involvedObject.name (appended to field_selector). cluster: Named cluster to target (empty = default).
Inputs: cluster, namespace, field_selector, involved_object. Returns: result
list_gatewaysB
List Istio Gateways with selector, ports, servers count, and age.
Inputs: cluster, namespace. Returns: result
list_gitops_applicationsC
List ArgoCD Applications with sync/health status, repo, revision, destination, and age.
Inputs: cluster, namespace. Returns: result
list_gitops_clustersB
List ArgoCD registered clusters from Secrets with label argocd.argoproj.io/secret-type=cluster.
Inputs: cluster, namespace. Returns: result
list_groupsB
List OpenShift Groups with member count and up to 5 member names.
Inputs: cluster. Returns: result
list_hpasC
List HorizontalPodAutoscalers with min/max/current/desired replicas and metrics.
Inputs: cluster, namespace. Returns: result
list_image_streamsC
List ImageStreams with their internal Docker repository and tag count.
Inputs: cluster, namespace. Returns: result
list_image_stream_tagsC
List tags for an ImageStream with image digest and creation timestamp.
Inputs: image_stream (required), cluster, namespace. Returns: result
list_inference_servicesC
List InferenceServices with readiness, model format, storage URI, URL, and age.
Inputs: cluster, namespace. Returns: result
list_ingressesA
List Kubernetes Ingress objects with NAMESPACE, NAME, CLASS, HOSTS, ADDRESS, PORTS, and AGE.
Args: namespace: Namespace to query (empty = all namespaces). label_selector: Label selector to filter Ingresses. cluster: Named cluster to target (empty = default).
Inputs: cluster, namespace, label_selector. Returns: result
list_installed_operatorsB
List installed operators (ClusterServiceVersions) with display name, version, and phase.
Inputs: cluster, namespace. Returns: result
list_install_plansC
List InstallPlans with approval mode, approved status, phase, and CSV names.
Inputs: cluster, namespace. Returns: result
list_integration_test_scenariosC
List Konflux IntegrationTestScenarios.
Inputs: namespace (required), cluster, application_name. Returns: result
list_jobsA
List Jobs with NAMESPACE, NAME, COMPLETIONS, ACTIVE, SUCCEEDED, FAILED, and AGE.
Args: namespace: Namespace to query (empty = all namespaces). label_selector: Label selector to filter Jobs. cluster: Named cluster to target (empty = default).
Inputs: cluster, namespace, label_selector. Returns: result
list_konflux_applicationsC
List Konflux Applications in a workspace namespace.
Inputs: namespace (required), cluster. Returns: result
list_machine_autoscalersC
List MachineAutoscalers in openshift-machine-api with target MachineSet, min, max, and age.
Inputs: cluster. Returns: result
list_machine_config_poolsC
List MachineConfigPools with rendered config, machine counts, and status flags.
Inputs: cluster. Returns: result
list_machine_configsC
List MachineConfigs with role label (machineconfiguration.openshift.io/role), generation, and age.
Inputs: cluster, label_selector. Returns: result
list_machinesC
List Machines with phase, node ref, provider ID, and age.
Inputs: cluster, label_selector. Returns: result
list_machine_setsA
List MachineSets with desired/ready/available replica counts, instance type, and age.
Inputs: cluster. Returns: result
list_managed_cluster_addonsC
List ManagedClusterAddons on a specific managed cluster.
Inputs: cluster_name (required), cluster. Returns: result
list_managed_clustersB
List ACM ManagedClusters with status, OCP version, cloud provider, and region.
Inputs: cluster. Returns: result
list_manifest_worksB
List ManifestWorks deployed to a ManagedCluster (namespace=cluster_name).
Inputs: cluster_name (required), cluster. Returns: result
list_model_registriesC
List ModelRegistries with availability, REST port, gRPC port, and age.
Inputs: cluster, namespace. Returns: result
list_model_serversC
List ServingRuntimes and ClusterServingRuntimes with scope, formats, multi-model, and age.
Inputs: cluster, namespace. Returns: result
list_namespacesA
List all namespaces with STATUS, LABELS, and AGE.
Inputs: cluster, label_selector. Returns: result
list_network_policiesA
List NetworkPolicies with NAMESPACE, NAME, POD-SELECTOR, POLICY-TYPES, and AGE.
Args: namespace: Namespace to query (empty = all namespaces). cluster: Named cluster to target (empty = default).
Inputs: cluster, namespace. Returns: result
list_nodesC
List all nodes with ROLES, STATUS, OS image, KUBELET version, and AGE.
Inputs: cluster, label_selector. Returns: result
list_notebooksC
List Kubeflow Notebooks with state, image, CPU/memory requests, and age.
Inputs: cluster, namespace. Returns: result
list_operator_conditionsC
List OperatorConditions (operators.coreos.com/v2) with active condition summary.
Inputs: cluster, namespace. Returns: result
list_operator_groupsC
List OperatorGroups with their target namespaces and service account.
Inputs: cluster, namespace. Returns: result
list_peer_authenticationsC
List Istio PeerAuthentications with mTLS mode, selector, and age.
Inputs: cluster, namespace. Returns: result
list_pipeline_runsC
List Tekton PipelineRuns with pipeline name, status, duration, and age.
Inputs: cluster, namespace, label_selector. Returns: result
list_pipelinesB
List Tekton Pipelines with task count and age.
Inputs: cluster, namespace. Returns: result
list_placementsC
List ACM Placements with predicates and satisfaction status.
Inputs: cluster, namespace. Returns: result
list_podsA
List pods with NAMESPACE, NAME, STATUS, READY, RESTARTS, NODE, and AGE columns.
Args: namespace: Namespace to query (empty = all namespaces). label_selector: Label selector (e.g. "app=nginx"). field_selector: Field selector (e.g. "status.phase=Running"). cluster: Named cluster to target (empty = default).
Inputs: cluster, namespace, field_selector, label_selector. Returns: result
list_pod_security_violationsB
List events that indicate pod security or SCC violations (FailedCreate + security keywords).
Inputs: cluster, namespace. Returns: result
list_policiesC
List ACM Policies with remediation action and compliance status.
Inputs: cluster, namespace. Returns: result
list_prometheus_rulesC
List PrometheusRules with counts of alerting rules and recording rules per resource.
Inputs: cluster, namespace. Returns: result
list_pvcsB
List PersistentVolumeClaims with status, volume, capacity, access modes, and storage class.
Inputs: cluster, namespace, label_selector. Returns: result
list_pvsC
List PersistentVolumes with capacity, access modes, storage class, reclaim policy, status, and claim.
Inputs: cluster, label_selector. Returns: result
list_release_plansC
List Konflux ReleasePlans with target workspace.
Inputs: namespace (required), cluster, application_name. Returns: result
list_releasesC
List Konflux Releases with snapshot, status, and target workspace.
Inputs: namespace (required), cluster, application_name. Returns: result
list_role_bindingsA
List RoleBindings in a namespace with role reference and bound subjects.
Inputs: cluster, namespace. Returns: result
list_rolesB
List Roles in a namespace with the number of policy rules.
Inputs: cluster, namespace. Returns: result
list_routesA
List OpenShift Routes (route.openshift.io/v1) with NAMESPACE, NAME, HOST/PATH, TLS, SERVICE, and AGE.
Args: namespace: Namespace to query (empty = all namespaces). cluster: Named cluster to target (empty = default).
Inputs: cluster, namespace. Returns: result
list_sccsB
List all SecurityContextConstraints sorted by priority descending.
Inputs: cluster. Returns: result
list_secretsB
List Secrets showing type and key count only — secret values are never displayed.
Inputs: cluster, namespace, label_selector. Returns: result
list_service_accountsC
List ServiceAccounts with number of secrets and image pull secrets.
Inputs: cluster, namespace. Returns: result
list_service_entriesC
List Istio ServiceEntries with hosts, location, ports, and age.
Inputs: cluster, namespace. Returns: result
list_service_mesh_control_planesC
List ServiceMeshControlPlanes with generation, version, readiness, and age.
Inputs: cluster, namespace. Returns: result
list_service_mesh_membersC
List ServiceMeshMemberRolls and the namespaces enrolled in each mesh.
Inputs: cluster, namespace. Returns: result
list_servicesA
List Services with NAMESPACE, NAME, TYPE, CLUSTER-IP, EXTERNAL-IP, PORT(S), and AGE.
Args: namespace: Namespace to query (empty = all namespaces). label_selector: Label selector to filter Services. cluster: Named cluster to target (empty = default).
Inputs: cluster, namespace, label_selector. Returns: result
list_silencesB
List Alertmanager silences with ID, state, matchers, comment, creator, and end time.
Inputs: cluster. Returns: result
list_snapshotsC
List Konflux Snapshots with components and integration test status.
Inputs: namespace (required), cluster, application_name. Returns: result
list_statefulsetsA
List StatefulSets with NAMESPACE, NAME, READY, SERVICE, and AGE.
Args: namespace: Namespace to query (empty = all namespaces). label_selector: Label selector to filter StatefulSets. cluster: Named cluster to target (empty = default).
Inputs: cluster, namespace, label_selector. Returns: result
list_storage_classesC
List StorageClasses. The cluster default is marked with (default).
Inputs: cluster. Returns: result
list_subscriptionsC
List OLM Subscriptions with package, channel, source, current CSV, and age.
Inputs: cluster, namespace. Returns: result
list_task_runsC
List Tekton TaskRuns with task name, status, duration, and age.
Inputs: cluster, namespace, label_selector. Returns: result
list_tasksC
List Tekton Tasks with step count and age.
Inputs: cluster, namespace. Returns: result
list_trigger_templatesB
List Tekton TriggerTemplates with template count and age.
Inputs: cluster, namespace. Returns: result
list_usersB
List OpenShift users with their identities and group memberships.
Inputs: cluster. Returns: result
list_virtual_machine_instancesA
List running VirtualMachineInstances with node, IP, and phase.
Inputs: cluster, namespace. Returns: result
list_virtual_machinesC
List VirtualMachines with status, vCPU, memory, and age.
Inputs: cluster, namespace, label_selector. Returns: result
list_virtual_servicesA
List Istio VirtualServices with hosts, gateways, HTTP/TCP route counts, and age.
Inputs: cluster, namespace. Returns: result
list_vm_snapshotsC
List VirtualMachineSnapshots with phase, source VM, and ready status.
Inputs: cluster, namespace. Returns: result
list_volume_snapshotsB
List VolumeSnapshots (snapshot.storage.k8s.io/v1) with source PVC, class, and ready status.
Inputs: cluster, namespace. Returns: result
list_vpasB
List VerticalPodAutoscalers with update mode and target workload.
Inputs: cluster, namespace. Returns: result
live_migrate_vmC
Live migrate a VirtualMachineInstance to another node.
Inputs: name (required), namespace (required), cluster. Returns: result
pause_machine_config_poolC
Pause a MachineConfigPool — prevents config updates from rolling out to nodes.
Inputs: name (required), cluster. Returns: result
pause_virtual_machineC
Pause a running VirtualMachine using virtctl.
Inputs: name (required), namespace (required), cluster. Returns: result
query_prometheusA
Execute a PromQL instant query against Thanos Querier and return results as a table.
Inputs: expr (required), cluster. Returns: result
query_prometheus_rangeA
Execute a PromQL range query. start/end in ISO8601 or Unix timestamp; step e.g. '5m'.
Inputs: end (required), expr (required), start (required), step, +1 more. Returns: result
refresh_gitops_applicationB
Force-refresh an ArgoCD Application by patching the argocd.argoproj.io/refresh annotation.
Inputs: name (required), cluster, namespace. Returns: result
remove_scc_from_service_accountC
Revoke an SCC from a ServiceAccount via 'oc adm policy remove-scc-from-user'.
Inputs: scc_name (required), namespace (required), service_account (required), cluster. Returns: result
remove_user_from_groupA
Remove a user from an OpenShift Group (atomic — uses oc adm groups remove-users).
Inputs: username (required), group_name (required), cluster. Returns: result
restart_virtual_machineC
Restart a VirtualMachine using virtctl.
Inputs: name (required), namespace (required), cluster. Returns: result
restore_vm_snapshotC
Restore a VM from a VirtualMachineSnapshot.
Inputs: vm_name (required), namespace (required), restore_name (required), snapshot_name (required), +1 more. Returns: result
rollout_deployment_configC
Trigger a new rollout of an OpenShift DeploymentConfig (oc rollout latest dc/).
Args: name: DeploymentConfig name. namespace: Namespace (default: "default"). cluster: Named cluster to target (empty = default).
Inputs: name (required), cluster, namespace. Returns: result
rollout_restart_deploymentB
Trigger a rolling restart of a Deployment (equivalent to 'oc rollout restart').
Args: name: Deployment name. namespace: Namespace (default: "default"). cluster: Named cluster to target (empty = default).
Inputs: name (required), cluster, namespace. Returns: result
rollout_status_deploymentA
Show the rollout status of a Deployment (equivalent to 'oc rollout status').
Args: name: Deployment name. namespace: Namespace (default: "default"). cluster: Named cluster to target (empty = default).
Inputs: name (required), cluster, namespace. Returns: result
rollout_undo_deploymentA
Undo a Deployment rollout, optionally to a specific revision.
Args: name: Deployment name. namespace: Namespace (default: "default"). revision: Target revision number (0 = previous revision). cluster: Named cluster to target (empty = default).
Inputs: name (required), cluster, revision, namespace. Returns: result
run_oc_commandA
Escape hatch: run any oc command. args is a space-separated string of arguments. WARNING: Use carefully. Destructive commands should be confirmed first.
Inputs: args (required), cluster. Returns: result
scale_deploymentB
Scale a Deployment to the specified number of replicas.
Args: name: Deployment name. replicas: Desired replica count. namespace: Namespace (default: "default"). cluster: Named cluster to target (empty = default).
Inputs: name (required), replicas (required), cluster, namespace. Returns: result
scale_machine_setA
Scale a MachineSet to the given number of replicas by patching spec.replicas. WARNING: Scaling down will cause machines (and their nodes) to be deleted.
Inputs: name (required), replicas (required), cluster. Returns: result
scale_statefulsetA
Scale a StatefulSet to the specified number of replicas.
Args: name: StatefulSet name. replicas: Desired replica count. namespace: Namespace (default: "default"). cluster: Named cluster to target (empty = default).
Inputs: name (required), replicas (required), cluster, namespace. Returns: result
start_buildC
Start a new build from a BuildConfig. Optionally supply from_dir to use a local directory as the binary source.
Inputs: build_config_name (required), cluster, from_dir, namespace. Returns: result
start_notebookA
Start a stopped Notebook by removing the kubeflow-resource-stopped annotation.
Inputs: name (required), namespace (required), cluster. Returns: result
start_pipeline_runB
Start a Tekton PipelineRun. params is a comma-separated list of KEY=VALUE pairs.
Inputs: pipeline_name (required), params, cluster, namespace. Returns: result
start_virtual_machineB
Start a VirtualMachine by setting spec.running=true.
Inputs: name (required), namespace (required), cluster. Returns: result
stop_notebookC
Stop a running Notebook by setting the kubeflow-resource-stopped annotation.
Inputs: name (required), namespace (required), cluster. Returns: result
stop_virtual_machineB
Stop a VirtualMachine by setting spec.running=false.
Inputs: name (required), namespace (required), cluster. Returns: result
sync_gitops_applicationC
Trigger a sync on an ArgoCD Application by patching the operation field.
Inputs: name (required), prune, cluster, dry_run, +1 more. Returns: result
uncordon_nodeA
Uncordon a node to allow new pods to be scheduled on it again.
Inputs: name (required), cluster. Returns: result
unpause_machine_config_poolB
Unpause a MachineConfigPool — allows pending config updates to roll out to nodes.
Inputs: name (required), cluster. Returns: result
unpause_virtual_machineC
Unpause a paused VirtualMachine using virtctl.
Inputs: name (required), namespace (required), cluster. Returns: result
update_configmapC
Set or update a single key in a ConfigMap using strategic merge patch.
Inputs: key (required), name (required), value (required), namespace (required), +1 more. Returns: result
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 216 tools
Every tool name includes a specific action and resource type (e.g., list_pods, get_pod, scale_deployment), making each tool's purpose clear and distinct. Even with 216 tools, the consistent naming prevents ambiguity, and generic escape hatches like run_oc_command are explicitly labeled.
All tools follow a consistent verb_noun pattern (list_, get_, create_, delete_, scale_, rollout_*, etc.), with resource types as the noun. Deviations like apply_manifest and get_resource still adhere to the same structure, ensuring predictability.
216 tools is extreme and far exceeds any reasonable scope for an MCP server. Even for a comprehensive OpenShift management platform, this count is overwhelming and will likely confuse agents, making tool selection difficult. It falls into the 'extreme mismatch' category.
The tool set covers an extensive range of OpenShift domains including compute, networking, storage, security, operators, machine management, monitoring, service mesh, OpenShift AI, virtualization, GitOps, Tekton, and ACM. However, some resource types lack individual get/update/delete operations (e.g., Deployments have no get or delete), relying on generic tools like get_resource and delete_resource, creating minor gaps.
Maintenance
Related MCP Connectors
MCP server for your apps' tools and custom tools, plus hosted AI agents and approval-gated workflows
MCP-Native LLM Orchestration Agent
MCP-native AI SRE: ask what's broken in production, get a reviewed GitHub fix PR.
The Google GKE MCP server is a managed Model Context Protocol server that provides AI applications with tools to manage Google Kubernetes Engine (GKE) clusters and Kubernetes resources. It exposes a structured, discoverable interface that allows AI agents to interact with GKE and Kubernetes APIs, enabling them to inspect cluster configurations, retrieve Kubernetes resource YAMLs, monitor operations like cluster upgrades, diagnose issues, and optimize costs—all without needing to parse text output or use complex kubectl commands.
Related MCP Servers
- AlicenseNot gradedqualityCmaintenanceAn open source MCP server empowering SREs with intelligent observability, predictive analytics, and AI-driven automation across Kubernetes, OpenShift, and Tekton environments.22 PyPI11Apache 2.0
- AlicenseAqualityDmaintenanceAI-powered MCP server for enterprise OpenShift/Kubernetes cluster management, providing diagnostic tools, RAG knowledge retrieval, and autonomous remediation recommendations.9MIT
- FlicenseNot gradedqualityDmaintenanceAn MCP server exposing Kubernetes-style diagnostic tools to an LLM agent, with a safety approval gate for destructive actions, all backed by a mock cluster for local testing.-
- AlicenseAqualityCmaintenanceAn MCP server exposing 72 tools across 26 homelab services, enabling LLMs to monitor and manage infrastructure, media, storage, and networking with a single endpoint.16MIT