Solaire MCP
A standalone MCP server that exposes YYLX gpt-image-2 image generation as a generate_image tool, returning inline images directly in the conversation via stdio.
README
Solaire MCP
A standalone Model Context Protocol (MCP) server that exposes YYLX gpt-image-2 image generation as the generate_image tool.
The server uses stdio, has no runtime npm dependencies, and returns generated images as inline MCP image content so clients such as Codex can display them directly in the conversation.
Why a standalone MCP repository?
The earlier implementation was bundled inside the solaire-plugins Codex Plugin. A bundled MCP is controlled from the Plugin page, but that page is not a general-purpose environment-variable editor.
This repository installs the image server as a normal standalone MCP server. It therefore appears in Codex's regular MCP Servers / Servers settings, where the server command, arguments, and environment variables can be managed separately from the Plugin.
Do not enable both the Plugin-bundled server and this standalone server with the same name. Doing so may create duplicate generate_image tools.
Features
- MCP tool:
generate_image - Model:
gpt-image-2 - OpenAI-compatible image-generation endpoint
- Streaming response support with completed-image preference
- Inline image output
- Environment-variable configuration
- Optional macOS Keychain fallback
- No third-party runtime dependencies
Requirements
- Node.js 18 or newer
- A YYLX-compatible base URL and API key
- An MCP client that supports stdio servers, such as Codex
Install in Codex Desktop
-
Open Settings in Codex.
-
Open MCP Servers or Servers (the label can vary by Codex version).
-
Choose Add server.
-
Enter:
Field Value Name yylx-gpt-imageCommand npxArguments -yandgithub:hsolaire/solaire-mcpas two arguments -
Add these environment variables in the same server settings:
Variable Example YYLX_BASE_URLhttps://app.yylx.ioYYLX_API_KEYyour local API key -
Save the server, then restart Codex or start a new task.
Keep the API key in your local Codex settings. Never commit it or paste it into a chat.
Install with Codex CLI
Add the standalone server command:
codex mcp add yylx-gpt-image -- npx -y github:hsolaire/solaire-mcp
Then add the two environment variables through Codex Desktop's server settings, or edit your local ~/.codex/config.toml:
[mcp_servers.yylx-gpt-image]
command = "npx"
args = ["-y", "github:hsolaire/solaire-mcp"]
[mcp_servers.yylx-gpt-image.env]
YYLX_BASE_URL = "https://app.yylx.io"
YYLX_API_KEY = "replace-this-locally"
Codex also supports forwarding variables already present in the Codex process environment:
[mcp_servers.yylx-gpt-image]
command = "npx"
args = ["-y", "github:hsolaire/solaire-mcp"]
env_vars = ["YYLX_BASE_URL", "YYLX_API_KEY"]
Run directly from GitHub
npx -y github:hsolaire/solaire-mcp
For reproducible installations, pin a release tag:
npx -y github:hsolaire/solaire-mcp#v0.1.0
Corresponding Codex arguments:
args = ["-y", "github:hsolaire/solaire-mcp#v0.1.0"]
Run from a local clone
git clone https://github.com/hsolaire/solaire-mcp.git
cd solaire-mcp
npm test
node ./bin/solaire-mcp.mjs
A local Codex configuration can launch the cloned file directly:
[mcp_servers.yylx-gpt-image]
command = "node"
args = ["/absolute/path/to/solaire-mcp/bin/solaire-mcp.mjs"]
[mcp_servers.yylx-gpt-image.env]
YYLX_BASE_URL = "https://app.yylx.io"
YYLX_API_KEY = "replace-this-locally"
Optional macOS Keychain setup
Environment variables configured in Codex are the recommended standalone setup. As an alternative on macOS, a local clone can store the values in Login Keychain:
npm run setup:keychain -- --base-url https://app.yylx.io
The helper asks macOS Keychain to prompt for the API key, so the key is not added to shell history. Environment variables take priority over Keychain values.
Optional Keychain-related variables:
YYLX_KEYCHAIN_ACCOUNTYYLX_KEYCHAIN_BASE_URL_SERVICEYYLX_KEYCHAIN_API_KEY_SERVICE
Tool schema
generate_image
| Argument | Required | Values / default |
|---|---|---|
prompt |
yes | non-empty string |
size |
no | 1024x1024, 1536x1024, 1024x1536; default 1024x1024 |
quality |
no | low, medium, high, auto |
background |
no | transparent, opaque, auto |
output_format |
no | png, jpeg, webp; default png |
n |
no | integer from 1 to 4; default 1 |
The native API sizes do not include 4K. For exact dimensions such as 3840x2160, generate with the closest native aspect ratio (1536x1024 for landscape), then resize with a separate image-processing step.
Update
Without a tag pin, npx -y github:hsolaire/solaire-mcp resolves the repository's current default branch. Restart the MCP server after updating.
For stable deployments, pin a release tag and change it intentionally when upgrading.
Uninstall
codex mcp remove yylx-gpt-image
You can also remove the server from Codex Desktop's MCP server settings.
Troubleshooting
The server does not appear
- Confirm the command is
npx. - Confirm the arguments are two separate values:
-yandgithub:hsolaire/solaire-mcp. - Restart Codex or start a new task after saving the server.
- Run
codex mcp listto check discovery.
npx cannot download the repository
Check GitHub access, DNS/proxy settings, and that Node.js/npm are available:
node --version
npm --version
For restricted networks, clone the repository once and use the local-file configuration above.
Credentials are missing
Set both YYLX_BASE_URL and YYLX_API_KEY in the standalone MCP server's environment-variable section. Do not add them to the repository or .env files that might be committed.
Duplicate image tools
Disable either the old Plugin-bundled yylx-gpt-image MCP or this standalone MCP. Keep only one enabled.
Image generation returns an upstream error
An MCP server can initialize and list tools even when the remote image API or key is unavailable. Verify the base URL and credential locally. A successful tools/list is not proof that a real image request will succeed.
Development
npm test
npm run check
npm pack --dry-run
License
MIT. See LICENSE.
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。