How to generate app mascots in Claude Code

What this gives you

MascotLab is an MCP server, so a coding agent can create artwork the same way it runs any other tool. You describe a character once, and every later image is generated against that locked design rather than from the description again. The agent gets back a download URL and can write the file into your repository without you leaving the terminal.

That last part is the difference between this and pasting prompts into an image tool. The mascot on your empty state and the mascot on your error page are the same character, because both were generated from the same locked reference rather than from two similar sentences.

• One mascot, then props, poses and animations that stay on-model
• Runs in Claude Code, Cursor, and any other MCP client
• Generations are asynchronous: you get an id back immediately and poll it
• Exports are PNG and MP4, and commercial use is included

Connect it to Claude Code

Two commands. The first registers the server at user scope so it is available in every project, and the second opens the browser to authorise it against your account.

claude mcp add --transport http --scope user mascotlab https://mascotlab.io/mcp

claude mcp login mascotlab

Run both from a terminal rather than from the desktop app's MCP panel. The login step opens a browser window, and the panel is a different surface that does not carry the flow through.

Connect it to Cursor

Cursor reads a JSON config. Save this as ~/.cursor/mcp.json to use MascotLab everywhere, or as .cursor/mcp.json inside a single project.

{
  "mcpServers": {
    "mascotlab": {
      "url": "https://mascotlab.io/mcp"
    }
  }
}

Other MCP clients take the same URL. The server speaks streamable HTTP and authorises with OAuth, so anything implementing the current spec can connect without a local process or an API key in a dotfile.

Generating your first character

Ask in plain language. The agent picks the tools; you do not need to name them.

> Create a mascot for this app: a calm capybara for a meditation
  tool, flat vector, soft colours. Then give me a waving pose and
  save it to public/mascot/.

Describe the character, not the change

This is the one piece of prompt advice that matters, and it is not obvious. A mascot's description becomes its identity: it is sent as the character description in every pose and animation you generate afterwards. A mascot described as "a blue version of the other one" carries no appearance in words at all, so every later pose has only the reference image to work from and drifts. Describe what the character looks like, even when you are making a variant.

Generations are asynchronous

Every create tool returns a generation id straight away rather than blocking. Images take roughly ten to forty seconds and animations one to three minutes, so the agent polls with check_generation until it settles. This is deliberate: a tool call that held the connection open for two minutes would time out in most clients.

Building on what you have

Once a mascot exists, everything else references it. A pose can take a second mascot so two characters share one frame. An animation can be pinned at both ends, starting on one pose and finishing on another, or looped so it ends exactly where it began and repeats with no visible cut.

Ask for those directly and the agent will pass the right ids: "animate the waving pose so it loops seamlessly", or "put the capybara and the fox in one pose, chasing each other".

What it costs

Credits, and nothing is charged for a generation that fails. Mascots, props and poses cost the same as each other; animations cost more because video is two orders of magnitude more expensive to render, and they need an active subscription rather than credits alone. Ask the agent to call get_account and it will tell you the current balance and the exact per-generation cost.

Artwork you generate is yours, including for commercial use. Exports are free and unwatermarked.