Kubermatic Agent Skills

Claude Code skills for working with Kubermatic Developer Platform (KDP). Install one of these skills so Claude knows how to talk to your KDP environment — browse the service catalog, enable services, create resources, check status, etc.

Skills

kdp

For platform users who just want to get things done. Uses kubectl under the hood but hides all the internal terminology — no mention of APIBindings, APIExports, kcp, or workspaces. Claude handles the plumbing and talks to you in plain language.

Requires: kubectl pointing at a KDP workspace.

kdp-mcp

Uses the mcp-kdp server. Claude calls its MCP tools directly — no shell commands and no kubectl. The server covers both the platform layer (service catalog, workspaces) and the resource layer (create, read, update, delete), so no second Kubernetes MCP server is needed.

Requires: the MCP endpoint hosted by your platform, added to your .mcp.json as kdp. Nothing to install.

kdp-kubectl

For service providers and platform engineers who want to see the internals. Same kubectl workflows as kdp but uses the actual KCP terminology (APIBindings, APIExports, etc.).

Requires: kubectl pointing at a KDP workspace.

kdp-blueprints

For Blueprint authors who want to compose several KDP services into one reusable, publishable kind (e.g. a PostgresPair that provisions two databases, or a WebappStack of database + repo + app). Walks through authoring a BlueprintDefinition with a kro ResourceGraphDefinition, validating it, publishing it to the service catalog, and smoke-testing an instance. Uses real platform terminology and includes an RGD authoring reference plus a complete example.

Requires: kubectl pointing at a KDP workspace with blueprints.kdp.k8c.io and the services you want to compose already bound.

Installation

Copy whichever skill you need into your Claude skills directory:

# project-level (just this repo)
mkdir -p .claude/skills/
cp -r skills/kdp .claude/skills/

# global (all projects)
mkdir -p ~/.claude/skills/
cp -r skills/kdp ~/.claude/skills/

Replace kdp with kdp-mcp, kdp-kubectl, or kdp-blueprints if you want one of the other variants.

Setup

kdp / kdp-kubectl

1. Point your kubeconfig at a KDP workspace
2. Check it works: kubectl api-resources
3. Optionally install the kcp kubectl plugin for workspace navigation

kdp-mcp

No install and no kubeconfig. Get the URL from the dashboard under Connect an AI agent, then:

claude mcp add --transport http \
  --client-id kdp-mcp \
  --callback-port 19876 \
  kdp https://public.api.<your-kdp-domain>/mcp

Run /mcp and log in through the browser. Every call acts as you, with your permissions.

Don't change the callback port — the matching redirect URI is pre-registered with the identity provider, which exact-matches it. Don't pin oauth.scopes either: the gateway advertises the scopes it needs, and overriding them breaks authorization.

The server is stateless: every tool takes an explicit workspace path such as root:demo, so tell Claude which workspace you are working in.

Usage

Just ask Claude what you need:

> What services are available?
> Enable the database service
> Create a PostgreSQL database called orders-db with 20Gi storage
> What's the status of my database?
> What credentials do I need to connect?

Contributing

1. Fork this repo
2. Add or modify skills under skills/
3. Each skill needs a SKILL.md with YAML frontmatter and instructions
4. Open a PR