agentregistry

Build. Deploy. Discover.

One Kubernetes-native registry for MCP servers, agents, skills, and prompts. Curate the AI building blocks your team trusts, then deploy them into your cluster through kagent with one command.

$ helm install agentregistry oci://ghcr.io/agentregistry-dev/agentregistry/charts/agentregistry
Install on Kubernetes Get started See in action

MCP Tools & Servers

Pull servers from npm, PyPI, OCI, or a remote URL into a catalog your platform team controls, then deploy them as Kubernetes workloads.

  • Search the registry by name, publisher, or tool
  • Enrichment scores and verified-publisher signals on every entry
  • Deployed through kagent as native Kubernetes resources
~/registry — arctl
❯ arctl get mcps
NAME TAG DESCRIPTION
k8s-cluster-inspector latest Query live cluster state — pods, deployments...
k8s-helm-release-mgr latest List, diff, rollback, and explain Helm releases...
❯ arctl apply -f k8s-cluster-inspector/deployment.yaml
✓ Deployment/k8s-cluster-inspector created
❯ arctl wait deployment k8s-cluster-inspector
✓ deployment k8s-cluster-inspector reached deployed on kubernetes-default
❯

Skills

Skills are versioned artifacts in the same catalog as everything else. Scaffold one, publish it, and anyone can pull it by name.

  • Scaffold with arctl init skill
  • Publish with a single arctl apply
  • Pull into a repo, or bundle into an agent blueprint
~/team-skills — arctl
❯ arctl init skill k8s-runbook
✓ Created skill: k8s-runbook
❯ arctl apply -f k8s-runbook/skill.yaml
✓ Skill/k8s-runbook (latest) created
❯ arctl get skills
NAME TAG DESCRIPTION
k8s-runbook latest Playbook for CrashLoopBackOff, ImagePullBackOff...
k8s-yaml-review latest Review Deployments, HPA, PDB, and NetworkPolicy...
❯ arctl pull skill k8s-yaml-review ./.claude/skills
✓ pulled skill/k8s-yaml-review (latest) into ./.claude/skills
❯

Agents

A blueprint records what an agent is — its identity, model, servers, skills, and prompts. Deploy it and kagent materialises it as native Kubernetes resources.

  • Build, push, and version blueprints with arctl
  • One registry, many connected clusters
  • Approve before deploy, with publisher and trust signals
~/agents — arctl
❯ arctl init agent k8sops --framework adk --language python
✓ Created agent: k8sops (framework: adk, language: python)
❯ arctl build ./k8sops --push
✓ pushed ghcr.io/acme/k8sops:latest
❯ arctl apply -f k8sops/agent.yaml -f k8sops/deployment.yaml
✓ Agent/k8sops (latest) created
✓ Deployment/k8sops-agent created
❯ arctl get deployments
NAME TARGET VERSION TYPE RUNTIME STATUS
default/k8sops-agent k8sops latest agent kubernetes-default deployed
❯

Prompts

Publish a prompt once and every agent referencing it picks up the pinned tag. Review the change before it ships.

  • Tagged versions, pinned per agent
  • See which agents consume each version
  • Served to clients through the registry MCP server
~/prompts — arctl
❯ arctl init prompt k8s-sre-operator --description "Careful Kubernetes operator"
✓ Created prompt: k8s-sre-operator
❯ arctl apply -f k8s-sre-operator.yaml
✓ Prompt/k8s-sre-operator (latest) created
❯ arctl get prompts
NAME TAG DESCRIPTION
k8s-sre-operator latest System prompt for a careful Kubernetes operator...
k8s-platform-engineer latest System prompt for Helm, GitOps, and golden-path...
❯ arctl get prompt k8s-sre-operator -o yaml
apiVersion: ar.dev/v1alpha1
kind: Prompt
❯

A catalog your developers want to browse.

Browse and search the catalog, manage artifacts, and deploy agents and MCP servers into a connected cluster — without leaving the browser.

registry.acme.dev / catalog / servers
The agentregistry web UI — Servers

Click around the window — tabs, rows, detail tabs, Deploy, and Add all work like the real UI, and the lists scroll.

Registry UI Browse the catalog, manage artifacts, deploy agents and MCP servers.
arctl CLI Scaffold, build, publish, pull, deploy — and script it all in CI/CD.
REST API The same capabilities as the UI and CLI, for your own portals.

Ship your first agent in five minutes.

Install the control plane with Helm, connect kagent as a runtime, then deploy an approved MCP server from the UI. Your IDE picks it up through the registry MCP server.

  1. 01
    Install on Kubernetes Helm installs the registry server and PostgreSQL into the agentregistry namespace.
  2. 02
    Connect a runtime Install kagent in the cluster. The built-in kubernetes-default runtime deploys agents and MCP servers through it.
  3. 03
    Open the UI and deploy Port-forward 12121, browse the catalog, deploy anything approved.
  4. 04
    Configure your IDE Run arctl configure cursor --port 31313 — or vscode, claude-code, kiro.
Read the full quickstart
~ — arctl
# 1. Install the control plane
❯ helm install agentregistry \
oci://ghcr.io/agentregistry-dev/agentregistry/charts/agentregistry \
--namespace agentregistry --create-namespace
✓ registry server ready · postgres ready
# 2. Verify the kagent-backed runtime
❯ arctl get runtimes
NAME TYPE
kubernetes-default Kubernetes
# 3. Open the UI and the registry MCP server
❯ kubectl port-forward -n agentregistry svc/agentregistry 12121 31313
ui: http://localhost:12121
mcp: http://localhost:31313
# 4. Wire it into your IDE
❯ arctl configure cursor --port 31313
✓ cursor configured for http://localhost:31313/mcp
❯

One control plane for every AI building block.

Install once, connect your clusters, and give every developer one place to discover and deploy.

Get started Join Discord