Get Started with agentregistry

Follow this guide to set up agentregistry and explore different scenarios for working with agents, MCP servers and skills.

Prerequisites

  • Docker and docker-compose
  • Claude code with Anthropic key set up (for the skill scenario)

Setup

Download and install the agentregistry CLI (arctl):

curl -fsSL https://raw.githubusercontent.com/agentregistry-dev/agentregistry/main/scripts/get-arctl | bash
1

Run an MCP server from the registry

In this demo you'll run an MCP server from a registry, locally on your computer.

If you try to list the MCP servers from the registry, you'll get an empty list because we haven't published any servers yet:

arctl mcp list -a

Let's publish an MCP server using the UI:

  1. Go to the UI and search for "io.github.Asthanaji05/pokeapi-mcp-server"
  2. Open the server details by clicking on the pokeapi-mcp-server card
  3. Make sure 1.12.0 is selected as the version
  4. Click on the Publish button to publish the server and make it available in the registry

Back in the CLI, you can now run arctl mcp list -a to see the server we published. Let's pull the server from the registry and run it locally:

arctl mcp run io.github.Asthanaji05/pokeapi-mcp-server --version 1.12.0 --inspector

You'll get prompted whether to run the latest version - press enter to continue. The server will start locally and the MCP inspector will automatically launch. You can connect to the server (select the "Streamable HTTP" transport type), list the tools and invoke them.

Try it out: Use the getPokemonByName tool, use "ditto" for the Pokemon name and check you get a response back.

Press CTRL+C to stop running the server and MCP inspector.

2

Add an MCP server to the registry

In this section we'll use the UI to add an MCP server to the registry. We will use the @modelcontextprotocol/server-everything as an example server to add.

  1. Open the UI
  2. Click the Add dropdown and select Add Server
  3. Fill out the server information:
    • Server name: "mycompany.com/server-everything"
    • Display title: "server everything"
    • Version: "0.0.1"
    • Description: "Server everything example entry"
  4. Click the Add Package button:
    • Package identifier: @modelcontextprotocol/server-everything
    • Version: 2025.9.25
  5. Click Create Server

The server will be added to your registry – you can find it in the list and click Publish to publish it. You'll see it in the "Published" tab in the UI or by running the arctl mcp list -a command.

3

Deploy an MCP server and connect to Claude code

You can deploy a published MCP server from the registry UI. Let's deploy the Pokemon and server-everything servers.

  1. Open the UI
  2. Click on the Published tab
  3. Click the Deploy button on the pokeapi-mcp-server and the server-everything entries

You'll see the deployed servers in the "Deployed" tab. Servers deployed in the registry have a well-known port they are accessible on, so they can be used in clients like VS Code, Cursor, Claude code and so on.

For example, you can run the following command in your project opened in Cursor:

arctl configure cursor --url http://localhost:21212/mcp

This command configures Cursor to connect to all MCP servers running in the registry. You'll be prompted by Cursor to enable the MCP server. You can click Enable and then open the "Tools & MCP" tab in the Cursor Settings to see the discovered tools and other resources from the MCP server.

4

Publish and pull a skill, then use it in Claude Code

Next, we'll deploy an existing skill to the agentregistry from the UI. We'll deploy an internal communications skill from Anthropic's skills demo repository.

  1. From the admin tab, click the Add dropdown and select Skill
  2. Enter the skill information:
    • Name: "my-internal-comms-skill"
    • Description: "My internal communications skill"
    • Version: "latest"
    • Docker image: "docker.io/pj3677/internal-comms"
  3. Click Add Skill
  4. Switch to the "Skills" list in the UI
  5. Open the "my-internal-comms-skill" details page
  6. Click on "Publish"

From the CLI you can check that the skill now shows up:

arctl skill list -a

Let's pull the skill:

arctl skill pull my-internal-comms-skill ~/.claude/skills/my-internal-comms-skill

We're pulling the skill directly to the ~/.claude/skills folder which makes the skill automatically available to Claude Code.

You can now launch Claude with claude and say something like:

"can you help me write a company newsletter"

Claude will prompt you before using the "internal-comms" skill. You can select Yes to continue and have Claude go through the skill instructions and help you create a company newsletter.

Next Steps

Join the agentregistry community and contribute to the future of AI development.