One registry for MCP servers, agents, skills, and prompts. Curate the AI building blocks your team trusts, deploy them with one command, and wire them into every IDE.
Agents, MCP tools, skills, and prompts — all curated in one registry your team can search, version, and deploy with a single command.
Identity, servers, skills, and prompts bundled into one deployable unit. Build, push, and version with arctl.
A built-in web UI to build, deploy, and create agents, servers, skills, and prompts — search the registry, kick off deployments, and track what's running, all without leaving the browser.
Platform teams stay in control of what's approved; developers get one place to discover and deploy. The same registry powers both.
Package, collect, and enrich AI artifacts from any source — into a single registry your whole company can trust.
Scaffold, test, publish, and deploy AI artifacts with minimal dependencies and no manual IDE wiring.
arctl init agent or arctl init skillarctl applyOne-line install. arctl drops into your $PATH and starts the local daemon on first run.
Any arctl command boots the registry. The web UI is exposed automatically.
Visit localhost:12121 and deploy any approved server or agent. Your IDE picks it up through the gateway.
Run arctl configure cursor (or vscode, claude) and you're done. New deployments are picked up automatically.
Instead of exposing each MCP server, agentregistry routes everything through agentgateway — an AI-native reverse proxy. Your clients connect to one URL, gateway enforces auth, and new deployments are picked up automatically.
Claude Code
OpenCode
arctl CLI
Any MCP client
agentregistry
An open-source platform that gives your team one place to find, manage, and run MCP servers, AI agents, skills, and prompts. You import or publish artifacts once, and anyone on the team can discover them, deploy them with one command, and have their IDE configured automatically.
agentregistry doesn't replace those — it sits on top of them. It pulls MCP servers from npm, PyPI, OCI, or remote endpoints into a single curated catalog with versioning, enrichment scores, verified publishers, and a one-command deploy path. Plus it ships with the agentgateway integration that makes those artifacts actually usable from Claude, Cursor, and VS Code.
Both — the workflow is identical. Locally, arctl brings up the daemon, web UI, and gateway via Docker Compose. For shared environments, deploy the same components into Kubernetes with the Helm chart. The same registry data and deployment YAML work in both places.
A blueprint is the versioned record of what an agent is — its identity, the MCP servers it needs, the skills it uses, and how it should be configured. A deployment is a running instance of that blueprint, scheduled into a specific environment with credentials and gateway routing wired up.
Yes. You can register servers from any internal OCI registry, npm proxy, private repo, or HTTP endpoint. The same governance and gateway flow applies — no public publishing required.
Yes — Apache 2.0. The CLI, daemon, web UI, and Helm chart are all in the agentregistry-dev/agentregistry repo. Contributions, issues, and feature requests are welcome.