Aurixel MCP – Geração de Imagens
bifrost-studio/aurixel-mcp
Geração de imagens IA integrada ao seu assistente Claude
Ver no GitHubServidor MCP que integra a plataforma Aurixel de geração de imagens (joyviz.ai) ao Claude Desktop, Cursor e outros hosts MCP. Permite gerar imagens sob demanda através de comandos de texto, com suporte a múltiplos modelos e resolução inteligente de timeouts via polling.
Instalação
npx convez-ai bifrost-studio/aurixel-mcp1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
Casos de uso
- Gerar ilustrações, artes visuais e designs diretamente do chat
- Criar mockups e conceitos visuais durante desenvolvimento
- Produzir assets de marketing e conteúdo visual sob demanda
Veredito

MCP bem arquitetado que resolve elegantemente o problema de timeouts longos via padrão job+poll. Ideal para fluxos que precisam de geração visual integrada, com setup direto e billing transparente via Aurixel.
README
A tradução para o português deste README ainda está na fila.
aurixel-mcp
An MCP server that lets MCP hosts
(Claude Desktop, Cursor, Cline, …) generate images through
Aurixel (joyviz.ai). It exposes Aurixel's image models — e.g.
gpt-image-2 — as tools, so you can just ask your assistant to draw
something and get the picture back.
Tools
| Tool | What it does |
|---|---|
generate_image(prompt, model?, size?) | Start a generation. Returns a job_id immediately (the image runs in the background). |
get_image_result(job_id) | Fetch the finished image. If it's still generating, returns a "still generating" message — call again with the same job_id. |
list_image_models() | List the image models you can pass to generate_image. |
Why a job + poll instead of returning the image inline?
gpt-image-2 takes 60–120s+, but MCP hosts cancel any single tool
call at a fixed ~60s timeout that progress notifications don't reliably
reset. So a long generation returned inline gets killed (and billed)
mid-flight. Instead, generate_image returns instantly and the host
polls get_image_result — each poll waits ≤30s, comfortably under the
host's limit, so generations of any length come through. No host-side
timeout tweak required. The assistant drives the polling for you; you
just ask for a picture and the image arrives when it's ready.
Setup
- Get a
ck-…API key at https://app.joyviz.ai/app/keys. - Add the server to your MCP host config.
Claude Desktop (claude_desktop_config.json) / Cursor
(~/.cursor/mcp.json):
{
"mcpServers": {
"aurixel": {
"command": "npx",
"args": ["-y", "github:bifrost-studio/aurixel-mcp"],
"env": { "AURIXEL_API_KEY": "ck-your-key-here" }
}
}
}
Restart the host, then ask it to "draw a watercolor fox" — it'll call
generate_image, poll get_image_result while the image renders, and
show the picture once it's ready.
Config (env)
| Var | Required | Default |
|---|---|---|
AURIXEL_API_KEY | ✅ | — |
AURIXEL_BASE_URL | https://conduit-api.joyviz.ai | |
AURIXEL_IMAGE_MODEL | gpt-image-2 |
Notes
- Transport: stdio (standard for locally-run MCP servers).
- Billing follows your Aurixel account — each image is charged to the key you configure.