xtool-mcp
Provides tools and resources for interacting with xTool laser machines, enabling material settings recommendations, design generation, troubleshooting, and machine knowledge through natural language.
README
xtool-mcp
An unofficial Model Context Protocol server, agent, and llms.txt for xTool laser machines — so you can bring an xTool-savvy assistant into your design studio and the AI client of your choice.
⚠️ Unofficial & safety notice. Not affiliated with, endorsed by, or sponsored by xTool. Everything here — settings, generated
.xcsfiles, project designs — is guidance, not a guarantee. Always test on scrap. Never cut PVC/vinyl or chlorine-containing materials. Never run a laser unattended. See DISCLAIMER.md.
What it does
- Machine knowledge — specs, work area, laser type, and supported software for the current xTool lineup (D1 family, S1, M1/M1 Ultra, F1/F1 Ultra/F2 series, P2/P3, MetalFab).
- Material settings — power/speed/passes/DPI starting points with provenance, confidence, and safety caveats — plus a "run a test grid" path when there's no good data.
- Software & troubleshooting — XCS vs Studio workflow help and symptom → cause → fix guidance.
- Design generation — produce cut-ready
.xcs(and SVG): material test grids, finger-joint boxes, and SVG → .xcs conversion. - Runs anywhere — one shared core, two entry points: a local stdio server (
npx xtool-mcp) and a remote Cloudflare Worker (Streamable HTTP).
Quick start
Use the hosted remote server
Add this Streamable HTTP endpoint to any MCP client:
https://xtool-mcp.doublexl.workers.dev/mcp
Run locally (stdio)
{
"mcpServers": {
"xtool": { "command": "npx", "args": ["-y", "xtool-mcp"] }
}
}
Drop that into Claude Desktop's claude_desktop_config.json or Cursor's mcp.json
(see agent/clients/), and paste agent/system-prompt.md
into your client's system-prompt field for the full "xTool laser assistant" persona.
What's inside
Tools
| Tool | Purpose |
|---|---|
list_machines / get_machine_specs |
Browse the machine catalog |
recommend_settings |
Power/speed/passes/DPI for a machine + material + thickness (with confidence + safety) |
search_knowledge |
Semantic/keyword search across machines, guides, troubleshooting, safety |
troubleshoot |
Likely causes and ordered fixes for a symptom |
generate_test_grid |
A .xcs/SVG grid varying two parameters (power vs speed) |
create_box |
Finger-joint box/tray panels as .xcs/SVG |
svg_to_xcs |
Convert an SVG into a cut-ready .xcs |
validate_xcs |
Check a .xcs for common structural problems |
Resources
xtool://machines, xtool://machines/{id}, xtool://materials,
xtool://troubleshooting, xtool://guides/{id}, xtool://safety, xtool://xcs-format.
Prompts
dial_in_settings, design_finger_joint_box, convert_svg_to_project,
troubleshoot_issue, plan_beginner_project, material_safety_check.
The .xcs format
.xcs is xTool's project file. It's community-reverse-engineered (plain JSON;
shapes in canvas[0].displays; laser parameters in a parallel device.data Map keyed
by shape id). This project generates the well-trodden single-file JSON used by D1/F1-class
machines. Open generated files in Studio/XCS and verify before running. Full writeup:
docs/xcs-format.md.
Development
Requires Node ≥ 18 and pnpm.
pnpm install
pnpm build # build core + stdio
pnpm typecheck
pnpm test
pnpm start:stdio # run the local MCP server
pnpm dev:worker # run the Cloudflare Worker locally
pnpm build:llms # regenerate llms.txt / llms-full.txt
pnpm build:embeddings # (optional) precompute the Workers AI search index
Architecture: docs/architecture.md ·
Data model & how to contribute settings: docs/data-model.md,
CONTRIBUTING.md.
Deploy the Worker
cd packages/worker
pnpm dlx wrangler deploy # deploys to xtool-mcp.<account>.workers.dev with an AI binding
The Worker also serves /llms.txt
and /llms-full.txt, generated live from the knowledge base.
Contributing
Material settings especially benefit from real-world data — see CONTRIBUTING.md. Every settings row needs a source and an honest confidence level.
License
MIT. "xTool", "xTool Creative Space", and "xTool Studio" are trademarks of their respective owners, used here only to describe compatibility.
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。