Skip to main content
Integrations & MCP Servers

Helm MCP server

The Helm MCP Server is a Model Context Protocol server that gives AI agents schema-validated control over Kubernetes Helm chart lifecycles. It exposes 18 tools, read-only cluster resources, and guided operational prompts so agents can search chart repositories, validate values.yaml overrides against JSON Schemas, run server-side dry-runs, monitor rollouts, and execute rollbacks safely.

18 Lifecycle Tools

Search ArtifactHub and custom repos, render templates, validate schemas, install/upgrade releases, and roll back revisions.

Dry-Run & Write Guardrails

Enforce read-only exploration with MCP_ALLOW_WRITE=false or require server-side dry-runs before applying changes.

Resources & Prompts

Read live release inventories (helm://releases), chart READMEs, and cluster readiness via MCP resources.

Workflows & Examples

Step-by-step patterns for chart discovery, pre-deployment validation, production upgrades, and rollbacks.

Core capabilities​

Capability DomainWhat the Server Provides
Chart Discovery & MetadataSearch charts across Bitnami, ArtifactHub, and private registries; inspect versions, READMEs, and values.schema.json definitions.
Pre-Deployment ValidationValidate custom values.yaml payloads against chart schemas, render manifests via helm template, and check cluster prerequisites before installation.
Release LifecycleInstall, upgrade, rollback, and uninstall Helm releases with built-in dry_run=True preview modes and revision history tracking.
Rollout MonitoringWatch pod readiness, release status, and Kubernetes warning events immediately after an install or upgrade.
Multi-Cluster ContextsEnumerate and switch Kubernetes contexts defined in KUBECONFIG with strict namespace isolation.

Architecture​

Quickstart​

1
Verify prerequisites

Ensure you have Docker (or Python 3.12+ with helm v3 and kubectl installed) and a valid ~/.kube/config pointing to your target cluster.

2
Start the Helm MCP Server container

Mount your kubeconfig read-only and expose port 8765:

docker run --rm -it \
  -p 8765:8765 \
  -v ~/.kube/config:/app/.kube/config:ro \
  talkopsai/helm-mcp-server:latest

The server listens on http://localhost:8765/mcp.

3
Register the server in your MCP client

Add the endpoint to your MCP client configuration (.cursor/mcp.json, claude_desktop_config.json, or OpsCode config.toml):

mcp.json
{
  "mcpServers": {
    "helm-mcp-server": {
      "url": "http://localhost:8765/mcp",
      "description": "Helm MCP Server for managing Kubernetes workloads via Helm"
    }
  }
}

Security and write protection​

  • Read-only mode: Set MCP_ALLOW_WRITE=false when you want agents to search charts, validate values, and run dry-run plans without permitting live cluster mutations.
  • Secret management: Never pass plaintext passwords or API keys in inline values payloads—reference existing Kubernetes Secret objects or External Secrets Operator resources.
  • Namespace isolation: Scope service accounts and RBAC roles to specific target namespaces.

Project structure​

helm-mcp-server/
├── helm_mcp_server/           # Main package
│   ├── tools/                 # 18 MCP tools
│   │   ├── discovery/         # Chart search and metadata inspection
│   │   ├── installation/      # Install, upgrade, rollback, uninstall
│   │   ├── validation/        # Values schema and manifest validation
│   │   ├── kubernetes/        # Context and namespace operations
│   │   └── monitoring/        # Release rollout health monitoring
│   ├── resources/             # MCP resources (helm://, k8s://)
│   ├── prompts/               # Operational runbook prompts
│   ├── services/              # Helm CLI & Kubernetes API services
│   ├── server/                # FastMCP server initialization
│   └── config.py              # Environment configuration
├── tests/                     # Automated test suite
└── pyproject.toml             # Package metadata
On this page