image_studio_mcp
An MCP server that lets any MCP client generate and edit images using Image Studio API, returning results inline and saving PNGs to disk.
README
image_studio_mcp
A Model Context Protocol server that lets any MCP client — Claude Code, opencode, Claude Desktop, or your own — generate and edit images in-loop with Image Studio. Results come back inline (the model sees the image) and each PNG is saved to disk.
You need nothing but Node ≥20 and an API key: point your client at npx @ross_technologies/image_studio_mcp,
drop in your key, and go. No clone, no build, no local model — all rendering runs on the hosted Image
Studio API.
- 📦 npm:
@ross_technologies/image_studio_mcp· v1.0.1 - 🔑 Get an API key: https://studio.ross-developers.com
- 📖 Full docs: https://studio.ross-developers.com/agents/mcp/docs
Quick start
Add this to your MCP client config (e.g. .mcp.json), put in your key, and restart the client:
{
"mcpServers": {
"image-studio": {
"command": "npx",
"args": ["-y", "@ross_technologies/image_studio_mcp"],
"env": { "IMAGE_STUDIO_API_KEY": "isk_your_key_here" }
}
}
}
Don't have a key yet? Use the watermarked, no-GPU sandbox key isk_test_onelove to try it out.
Then just ask your agent to make an image — e.g. "generate a sunset over Kingston harbour" — and
it will call generate_image, show you the result inline, and save the PNG.
Tools
| tool | description |
|---|---|
list_models |
list the image models (generation + editing) with their tasks and parameters |
generate_image |
text→image (irie, vibes, roots, sunsplash, criss, kaya, wagwan) — blocks, returns inline + saves |
edit_image |
image edit (remix, bashment, patchie[mask], bredda, rocksteady, biggup[upscale]) — input_image = path / data URI / base64 |
get_job |
poll a job by id |
cancel_job |
cancel a job |
generate_image / edit_image submit the job, poll to completion (up to MCP_MAX_WAIT_S), return
every image as an inline base64 block and write it to IMAGE_OUTPUT_DIR/<job_id>/<n>.png. Call
list_models first to see every model, its task, and its tunable parameters.
Configuration
Set via your MCP client's env block (above) or a .env file. See .env.example.
| var | required | default | notes |
|---|---|---|---|
IMAGE_STUDIO_API_KEY |
✅ | — | Your key from https://studio.ross-developers.com. Server refuses to start without it. |
IMAGE_OUTPUT_DIR |
<package>/outputs |
where PNGs are saved | |
MCP_MAX_WAIT_S |
900 |
max seconds a tool blocks before returning a job handle |
Install (optional)
npx needs no install. To pin it globally instead:
npm install -g @ross_technologies/image_studio_mcp
# then use "command": "image_studio_mcp" (no npx) in your MCP config — the bin is unscoped
From GitHub Packages
Also mirrored to GitHub Packages under the @ross-sec scope. Point that scope at GitHub's registry, then install:
echo "@ross-sec:registry=https://npm.pkg.github.com" >> .npmrc
npm install @ross-sec/image_studio_mcp
Releases
- Latest: v1.0.1 — see CHANGELOG.md.
- npm: https://www.npmjs.com/package/@ross_technologies/image_studio_mcp
- GitHub Packages: https://github.com/ross-sec/image_studio_mcp/packages
How it works
generate_image / edit_image call your Image Studio backend, poll the job to completion, and return
the finished PNG both inline (so the model sees it) and on disk. Nothing runs locally — the server
is a thin, typed client over the hosted /v1 API. It talks to your MCP client over stdio, and logs
only to stderr (stdout is the JSON-RPC stream). Your IMAGE_STUDIO_API_KEY is read from the client
env / a local .env and is never written to disk or logged. See SECURITY.md.
Contributing & conduct
Contributions welcome — see CONTRIBUTING.md. This project follows the Contributor Covenant.
License
MIT © Andre Ross
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。