Guides

MCP server

Connect Rogue to Claude Code, Claude Desktop, Cursor or any other MCP client, and ask it to create images and videos for you. The tools use the same keys, scopes, budgets and queue as the REST API.

Endpoint

The server speaks Streamable HTTP at one URL and authenticates with an API key in the Authorization header, exactly like the REST API:

MCP URL

https://rogue-api.sentryq-va.com/mcp

Use a dedicated key

Create a key just for your agent on the API keys page, with a daily credit budget. The console shows every snippet below with the key already filled in.

Claude Code

Run this once in a terminal. The Rogue tools are available in your next session.

Terminal

claude mcp add --transport http rogue https://rogue-api.sentryq-va.com/mcp --header "Authorization: Bearer rk_live_YOUR_KEY"

Add --scope user to make it available in every project, and check the connection with claude mcp list.

Claude Desktop

Open Settings, then Developer, then Edit Config, and add the server to claude_desktop_config.json. Restart Claude Desktop afterwards.

claude_desktop_config.json

{
  "mcpServers": {
    "rogue": {
      "url": "https://rogue-api.sentryq-va.com/mcp",
      "headers": {
        "Authorization": "Bearer rk_live_YOUR_KEY"
      }
    }
  }
}

If your version only runs local servers

Some Claude Desktop versions only start local, command-based servers from this file. Bridge to the remote server with mcp-remote (needs Node.js):

claude_desktop_config.json

{
  "mcpServers": {
    "rogue": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-remote",
        "https://rogue-api.sentryq-va.com/mcp",
        "--header",
        "Authorization:${ROGUE_AUTH}"
      ],
      "env": {
        "ROGUE_AUTH": "Bearer rk_live_YOUR_KEY"
      }
    }
  }
}

Cursor

Add the server to ~/.cursor/mcp.json for every project, or to .cursor/mcp.json in one project. Cursor picks it up under Settings, MCP.

mcp.json

{
  "mcpServers": {
    "rogue": {
      "url": "https://rogue-api.sentryq-va.com/mcp",
      "headers": {
        "Authorization": "Bearer rk_live_YOUR_KEY"
      }
    }
  }
}

Other clients

Any client that supports remote servers over Streamable HTTP with custom headers works: point it at https://rogue-api.sentryq-va.com/mcp and send Authorization: Bearer rk_live_... (or X-API-Key).

Tools

Tools a key lacks the scope for answer with a clear error, so give an agent only the scopes it needs.

Tool	Scope	What it does
`get_account`	`read`	The Rogue account, its plan and credit balance.
`list_models`	`read`	Models for a suite (image, edit, video, extend) with the settings each accepts.
`list_styles`	`read`	Style ids to pass in styles (up to three).
`list_suggestions`	`read`	Prompt suggestions such as outfits, locations and lighting.
`list_projects`	`read`	Projects on the account.
`create_project`	`projects`	Creates a project to group generations.
`upload_from_url`	`uploads`	Imports an image, video or audio file from an https URL.
`estimate_cost`	`generate`	Prices a generation in credits before running it.
`generate_image`	`generate`	Creates images from a prompt. Spends credits.
`edit_image`	`generate`	Transforms or inpaints an existing image. Spends credits.
`generate_video`	`generate`	Creates a video from a prompt and optional frames. Spends credits.
`extend_video`	`generate`	Continues an existing video. Spends credits.
`get_generation`	`read`	Waits for a generation to finish and returns its outputs.
`list_generations`	`read`	Recent generations with their status.
`search_library`	`read`	Searches images and videos in the Rogue library.

Generation tools accept wait_seconds (up to 60) to return the finished result in the same call, and get_generation waits for a job that is still running, so an agent needs no polling loop of its own.

Credits and safety

Spending tools say so in their description, so clients can ask before running them. Price a request first with estimate_cost.
A key's daily credit budget applies to its agent too: past the budget, generation tools fail instead of spending.
Adult media is never returned inline to the agent. Results carry URLs and metadata only; open them yourself, or look in Generations, where mature outputs stay blurred until you reveal them.