ktl - K8s manifest generator, transformer, and analyzer

ktl is a versatile CLI tool for working with Kubernetes manifests. It can pull resource definitions from live clusters or static files, apply various transformations, and output results in various formats like Kustomize, Helm, CSV and more.

Think of it as your Swiss Army knife for:

• Migrating live cluster resources to deployment & GitOps tools like k0rdent, Flux, ArgoCD, and others
• Normalizing and converting manifests
• Querying and analyzing resources across multiple clusters
• Creating Model Context Protocol (MCP) tools to enrich LLM context

Usage Scenarios

Migration & manifest generation tool

Convert your live cluster state into reproducible Kustomize packages or Helm charts. Perfect for streamlining migration to services like k0rdent, Flux or ArgoCD.

$ ktl run generate.yaml

Example of generate.yaml to generate a Kustomize components package with all resources from the simple-app namespace, except for specific resources:

source:
  kind: KubeConfig
  clusters:
  - names: dev-*
    tags: dev
  - names: test-cluster-a,test-cluster-b
    tags: test
  - names:
    - 'prod-*'
    - '*-prod-*'
    tags: prod
  resources:
  - namespaces: simple-app
filters:
- kind: SkipFilter
  resources:
  - kind: CronJob
    name: infra-canary
output:
  kind: KustomizeComponents

If you prefer Helm chart, just change the output:

output:
  kind: HelmChart
  helmChart:
    name: simple-app
    version: v1.0

Manifest cleanup, conversion & optimization tool

$ ktl run convert.yaml

Normalize and optimize your existing manifests while also removing unwanted fields:

source:
  kind: Kustomize
  kustomization: "path-to/${CLUSTER}"
  clusters:
  - names: dev-*
    tags: dev
  - names: test-cluster-a,test-cluster-b
    tags: test
  - names:
    - prod-cluster-a
    - prod-cluster-b
    tags: prod
filters:
- kind: SkipFilter
  fields:
  - metadata.annotations.example\.com/generated
  - metadata.annotations.dev\.example\.com/info[=flaky-tests]
output:
  kind: KustomizeComponents

Or convert to a Helm chart with output:

output:
  kind: HelmChart
  helmChart:
    name: converted-app
    version: v1.0

Query and analysis tool

Use the query command to find matching resources:

$ ktl query 'deployments.apps,daemonsets' \
  --clusters 'dev-*' \
  'it["spec.template.spec.containers.[image=db:.*]"]'

Or use the run command for more advanced filtering using the Starlark Python dialect:

$ ktl run query.yaml

query.yaml:

source:
  kind: KubeConfig
  clusters:
  - names: dev-*
    tags: dev
  - names: test-cluster-a,test-cluster-b
    tags: test
  - names:
    - 'prod-*'
    - '*-prod-*'
    tags: prod
  resources:
  - namespaces: simple-app
filters:
- kind: Starlark
  script: |
    for it in resources:
      if not it["spec.template.spec.containers.[image=db:.*]"]:
        continue
      if "test" in str(it.metadata.name):
        continue
      output.append(res)
output:
  kind: Table
  columns:
  - name: CLUSTER
    text: "${CLUSTER}"
  - name: KIND
    field: kind
  - name: NAMESPACE
    field: metadata.namespace
  - name: NAME
    field: metadata.name
  - name: CONTAINER
    field: spec.template.spec.containers.*.name
  - name: IMAGE
    field: spec.template.spec.containers.*.image

MCP tool (proof-of-concept)

Create MCP tools that can provide your LLM with information about resources on your clusters. Combine with the above query filters for advanced scenarios.

NOTE: currently the MCP server is implemented as mcp-bridge-poc.py script, but the future versions will provide it as part of the ktl binary.

source:
  clusters:
  - names: "*"
  resources:
  - apiResources: deployments.apps
output:
  kind: MCPTool
  description: |-
    List containers and their images for
    all deployments in all known K8s clusters
  columns:
  - name: CLUSTER
    text: "${CLUSTER}"
    description: Cluster name
  - name: NAMESPACE
    field: metadata.namespace
    description: Namespace name
  - name: NAME
    field: metadata.name
    description: Deployment name
  - name: CONTAINER
    field: spec.template.spec.containers.*.name
    description: Container name
  - name: IMAGE
    field: spec.template.spec.containers.*.image
    description: Container image