Allows natural language Kubernetes operations including pod management, scaling, runbooks, and monitoring.
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., "@Kubernetes + Prometheus SRE MCP Server List crashlooping pods in production"
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.
Natural language Kubernetes operations — powered by Model Context Protocol (MCP)
Built to scale from a single cluster to multi-cluster, multi-team enterprise environments.
An MCP (Model Context Protocol) server that exposes Kubernetes SRE operations as tools an AI assistant can call.
You: "Run the high error rate runbook for the production namespace"
Claude: [calls run_runbook → executes org-approved diagnosis sequence]
Step 1: Checked deployments — nginx (3/3), api-service (1/3 ⚠️)
Step 2: Found pod api-service-7f9d — 47 restarts, OOMKilled
Step 3: Warning events — OOMKilled x3 in last 10 minutes
Recommendation: Increase memory limit to 512Mi + scale to 5 replicasRelated MCP server: claude-c2
Feature | v1 | v2 |
Clusters supported | 1 (hardcoded) | Many (dynamic context switching) |
Write operations | Unrestricted | Policy-checked with guardrails |
Audit trail | None | Full structured JSON log |
Incident diagnosis | Ad-hoc | Encoded runbooks (standardized) |
Operational consistency | Per-engineer | Org-wide enforced |
Tool | Description |
| All clusters in kubeconfig |
| Pod status, restarts, container states |
| CrashLoopBackOff pods across all namespaces |
| Logs including previous crashed container |
| Node readiness and pressure conditions |
| Desired vs ready vs available replicas |
| Warning events — key incident signal |
| All namespaces |
Tool | Policy Enforced |
| Max replicas · Blocked namespaces · Prod minimums |
Tool | Description |
| Available runbooks with triggers |
| Execute org-standard diagnosis sequence |
Tool | Description |
| All recent operations with timestamps |
Claude Desktop (MCP Host)
│
│ MCP Protocol (stdio / JSON-RPC)
▼
┌─────────────────────────────────────┐
│ SRE MCP Server v2 │
│ server.py ← entry point │
│ cluster_manager ← multi-cluster │
│ policy.py ← write guards │
│ audit.py ← JSON audit log │
│ runbooks.py ← SRE runbooks │
└──────────────┬──────────────────────┘
│ kubernetes Python SDK
▼
┌──────────────────────┐
│ Kubernetes Clusters │
│ (any kubeconfig │
│ context) │
└──────────────────────┘git clone https://github.com/ManishMaurya22/sre-mcp-server
cd sre-mcp-server
python3.11 -m venv venv
source venv/bin/activate
pip install -r requirements.txtEdit ~/Library/Application Support/Claude/claude_desktop_config.json:
{
"mcpServers": {
"sre-k8s": {
"command": "/Users/<YOUR_USERNAME>/sre-mcp-server/venv/bin/python",
"args": ["/Users/<YOUR_USERNAME>/sre-mcp-server/server.py"]
}
}
}See docs/SETUP.md for full setup guide.
export POLICY_MAX_REPLICAS=30
export POLICY_SCALE_BLOCKED_NS="kube-system,gatekeeper-system"
export POLICY_PROD_NAMESPACES="production,prod"
export POLICY_PROD_MIN_REPLICAS=2You: "Scale nginx to 0 in production"
Claude: ❌ Policy Denied — scaling to 0 not allowed in production (min: 2)
Operation audit-logged.Available: high_error_rate · node_pressure · deployment_rollback
You: "Run the high_error_rate runbook for production"
Claude runs in order:
1. get_deployments → spot unhealthy deployments
2. get_pods → check restart counts
3. get_events → surface warning signals
4. get_crashlooping_pods → cluster-wide check
+ surfaces remediation hintssre-mcp-server/
├── server.py # Main MCP server
├── cluster_manager.py # Multi-cluster context management
├── policy.py # Write operation guardrails
├── audit.py # Structured audit trail
├── runbooks.py # Encoded SRE runbooks
├── requirements.txt
├── tools/k8s_tools.py
├── config/claude_desktop_config.example.json
├── docs/
│ ├── SETUP.md
│ └── INTERVIEW_GUIDE.md
└── .github/workflows/ci.yamlPrometheus MCP — SLO burn rate queries
PagerDuty MCP — incident acknowledgement
ArgoCD MCP — GitOps sync and triggers
Central MCP Gateway — auth + multi-team routing
MIT — See LICENSE
Built by Manish Maurya — DevOps/SRE Leader | 16+ Years | Abu Dhabi, UAE Website: https://manishmaurya22.github.io/
This server cannot be deployed
Create projects, nodes, and tasks in UluP Spaces by conversation with Claude.
Persistent context for Claude. Your AI always knows your projects and next actions across sessions.
Deploy, monitor, and manage your OpenClaw AI assistants via natural language.
Monitoring + status pages set up by talking to Claude. Auto-detects 30+ SDKs and your URLs.