Skip to content

Setting Up Agent MCP

The Rogue Arena MCP server is an authenticated proxy that gives your AI agent direct access to your Rogue Arena account. Once it’s set up, your agent can read and change canvases, manage deployments, build curriculum, and develop plugins from your terminal.

It’s a standard MCP server, so it works with any agent that supports MCP: Claude Code, OpenAI Codex, Grok, and more. The installer sets up Claude Code and Codex for you; other agents take one line of config.

Source and full details: github.com/rogue-labs-cyber/rogue-arena-mcp

  • Node.js 20.9+
  • Git
  • An AI agent that supports MCP. For automatic setup, install Claude Code or OpenAI Codex first. The installer won’t install an agent for you.

macOS / Linux:

Terminal window
curl -fsSL https://raw.githubusercontent.com/rogue-labs-cyber/rogue-arena-mcp/main/install.sh | sh

Windows (PowerShell):

Terminal window
irm https://raw.githubusercontent.com/rogue-labs-cyber/rogue-arena-mcp/main/install.ps1 | iex

The installer builds the MCP server and installs the rogue-mcp command. Then it looks for Claude Code and Codex and sets up whichever it finds: it registers the MCP server with each one and installs the skills.

Terminal window
rogue-mcp login

A browser window opens for you to sign in with your Rogue Arena account. Your token is stored in your OS keychain and refreshes automatically for 31 days. One login covers every agent on the machine.

Restart your agent after logging in. Until you log in, the server only offers a single rogue_auth_status tool.

Terminal window
rogue-mcp whoami

It prints your username, the hub, the profile, where your credentials are stored, and when your session expires.

For any other MCP-capable agent, such as Grok, run the installer and log in as above, then add the server to your agent’s MCP settings. It runs the command rogue-mcp serve. Most agents accept a config like this; check your agent’s docs for where it goes:

{
"mcpServers": {
"rogue-arena": {
"command": "rogue-mcp",
"args": ["serve"]
}
}
}

Your agent gets every Rogue Arena tool. The skills below are set up automatically for Claude Code and Codex; with other agents, describe what you want and the agent works from the tools directly.

The rogue-mcp command connects your agent to arena.roguelabs.io and passes every tool call through with your sign-in. Your agent gets tools for:

  • Reading and changing canvases (machines, VLANs, plugins, forests)
  • Managing deployments (start, monitor, debug)
  • Uploading and downloading files to VMs
  • Working with curriculum content
  • Developing and testing Ansible plugins

Upload and download tools can read and write files on your computer. The server blocks sensitive folders such as ~/.ssh, ~/.aws, and ~/.gnupg.

Skills teach your agent the Rogue Arena workflows.

SkillWhat it unlocks
rogue-build-scenarioDesign and build lab scenarios: brainstorm, implement, validate, and debug
rogue-active-deploymentOperate on running deployments: run commands, browse files, test payloads
rogue-plugin-devDevelop Ansible plugins: brainstorm, write YAML, manage vaults
rogue-curriculum-builderBuild course content: chapters, CTF nodes, media, quiz questions

Running a skill. Run a skill by name. In Claude Code, type /rogue-active-deployment; that’s the form these docs show. In Codex, type $rogue-active-deployment, or /skills to browse them. You can also describe what you want and name the skill in the same message. Claude Code doesn’t start these skills on its own.

Terminal window
rogue-mcp login # Sign in to Rogue Arena (shared by every agent)
rogue-mcp logout # Sign out and revoke the token
rogue-mcp whoami # Show sign-in, hub, profile, credential store, and session expiry
rogue-mcp serve # Start the MCP server (your agent runs this for you)
VariableDefaultDescription
ROGUE_HUB_URLhttps://arena.roguelabs.ioHub API URL
ROGUE_VAULTS_URLSame as hubVaults service URL
ROGUE_CLIENT_IDrogue-mcpKeycloak client ID
ROGUE_DISABLE_UPDATE_CHECKunsetSet to 1 to silence the update notice in Claude Code
ROGUE_PROFILEdefaultProfile name for side-by-side installs, such as a dev profile. Each profile keeps its own sign-in
ROGUE_CREDENTIALS_FILEunsetAbsolute path to a credentials file to use instead of the OS keychain, for headless machines. The file is written with 0600 permissions

Run the installer again. It refreshes the MCP server, the rogue-mcp command, and the skills for every agent it finds, and keeps your sign-in. Claude Code also shows a notice when an update is available.

Terminal window
rogue-mcp logout
claude mcp remove --scope user rogue-arena # Claude Code
codex mcp remove rogue-arena # Codex
npm uninstall -g rogue-arena-mcp
rm -rf ~/.rogue-arena-mcp ~/.claude/plugins/cache/rogue-arena
rm -rf ~/.codex/skills/architect-* ~/.codex/skills/rogue-*

For another agent, remove the rogue-arena entry from its MCP settings.