Decky Plugin Studio
MCP server for Decky Loader plugin development, providing tools to build, deploy, debug, and capture screenshots/recordings on a Steam Deck via commands like deck.deploy and deck.captureScreenshot.
README
Decky Plugin Studio
VS Code / Cursor extension for Decky Loader plugin development: live in-IDE QAM preview, MCP tools for build/deploy/debug on a real Steam Deck, composited screenshots and screen recordings, and an agent pack for your plugin repo.
Live preview is very much beta. It is great for fast UI and RPC iteration, but focus, layout, Steam Input, and gamescope compositing still need on-device QA. See Preview limitations.
What you get
| Feature | What it does |
|---|---|
| Live preview | QAM-sized webview with HMR, virtual gamepad, hardware simulator, Python sidecar for real main.py RPC |
| MCP tools | Agents and commands for deck.deploy, deck.captureScreenshot, deck.record, preview suite, tunnel/ingest |
| Composited capture | Screenshots and recordings that include QAM + your plugin UI (not raw game-only kmsgrab) |
| Init Pack | Drops AGENTS.md, Cursor/VS Code MCP config, skills, and optional scripts/ into your plugin repo |
| Create New Plugin | Clones decky-plugin-template, renames boilerplate, runs Init Pack |
Install
- Download the latest
.vsixfrom GitHub Releases - Extensions → … → Install from VSIX…
- Reload the editor
The status bar (bottom-right) shows the installed version, preview/tunnel state, and ingest count.
Quick start
1. Open a plugin workspace
Any folder with plugin.json and main.py. This repo includes example-plugin/ for smoke testing.
2. Configure your Deck (remote dev)
Use MCP deck.configure or create .env in your plugin repo:
DECK_IP=192.168.x.x
DECK_USER=deck
Deck credentials are also stored under ~/.config/decky-plugin-studio/deck.env when using MCP configure.
3. Init Pack
Command Palette → Decky: Init Pack
Copies agent guidance, MCP config, skills, and optional scripts (record-deck, screenshot-deck, preview suite).
4. Open preview (beta)
Command Palette → Decky: Open Preview
- Edit
src/→ Vite HMR in the QAM frame - Edit
main.py→ Python sidecar restarts with preserved state
Do not ship based on preview alone — deploy to a Deck for focus and layout bugs.
5. Day-to-day loop
preview (fast UI) → plugin.build → deck.deploy → on-device QA
| Task | How |
|---|---|
| Build plugin zip | MCP plugin.build or your pnpm run build |
| Deploy to Deck | MCP deck.deploy or Decky: Deploy to Deck |
| Screenshot (QAM + plugin) | MCP deck.captureScreenshot — open QAM + plugin first |
| Screen recording | MCP deck.record — open QAM + plugin before/during capture |
| Debug logs from Deck | deck.startTunnel → deck.probeIngest / deck.tailIngest |
| Agent automation | See MCP tools |
Artifacts land in your plugin workspace: screenshots/, recordings/.
6. MCP in Cursor / VS Code
After Init Pack, your plugin repo’s mcp.json points at the extension’s MCP server. Tools are documented in docs/MCP_TOOLS.md.
Example agent flows:
deck.configure→plugin.build→deck.deploy→deck.captureScreenshotpreview.start→preview.runSequencefor smoke testsdeck.recordwith{ seconds: 15, mode: "game", quality: "compressed" }
Capture prerequisites
For game mode screenshots and recordings:
- Deploy your plugin (
deck.deploy) - Open QAM on the Deck
- Open your plugin panel and keep it visible during capture
Recordings use composited pipewire-gamescope (game) or wf-recorder (desktop). They fail closed if only game-plane capture would succeed — unless you pass allowNonPluginUi: true.
Optional: deck.installCaptureHelper installs studio-record / studio-capture to ~/.local/bin on the Deck.
Developing this extension
- Branch model: feature work on
develop→ merge tomainwith a version bump inextension/package.json→ CI publishes a GitHub Release + VSIX - Build locally: docs/DEVELOPMENT.md
More documentation
- MCP tools reference
- Preview limitations — beta preview vs on-device QA
- Consumer sync (bonsAI & other plugins)
- Device QA runbook
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。