mcp-glasswarp
Allow AI agents to see and control a real Windows PC you own: observe (UIA + screenshots), click/type/drag/scroll, launch apps, owner Live View. BYOH — your machine, your key.
README
Glasswarp MCP Server
See and control a real Windows PC you own — from any MCP client, locally or remotely. Model Context Protocol over the Glasswarp Platform API: observe (UIA + screenshots), click/type/drag/scroll, launch apps, owner Live View. Hosted MCP or npx @glasswarp/mcp. BYOH: your machine, your key; you bring the model (task logic).
| Remote URL | https://mcp.glasswarp.com/mcp |
| Auth | Authorization: Bearer gw_… (create a key) |
| Docs | docs.glasswarp.com/get-started/mcp |
| License | Apache-2.0 |
| Registry | Official MCP: com.glasswarp/mcp-server |
Open: this MCP server + the Python SDK. Not open: the Windows host agent and Glasswarp platform (gateway, console, billing).
Quick connect
Cursor (recommended) — npx @glasswarp/mcp
Local stdio bridge to the remote server (same tools; needs a paired rig + API key):
{
"mcpServers": {
"glasswarp": {
"command": "npx",
"args": ["-y", "@glasswarp/mcp"],
"env": {
"GLASSWARP_API_KEY": "gw_live_sk_REPLACE_WITH_YOUR_KEY"
}
}
}
}
If bare npx fails inside Cursor (ENOENT under Cursor.app), use an absolute path (e.g. /opt/homebrew/bin/npx) and set PATH accordingly.
Package source: sdk/mcp → npm @glasswarp/mcp.
Cursor — raw mcp-remote (equivalent)
{
"mcpServers": {
"glasswarp": {
"command": "/opt/homebrew/bin/npx",
"args": [
"-y",
"mcp-remote",
"https://mcp.glasswarp.com/mcp",
"--header",
"Authorization:${GLASSWARP_AUTH}"
],
"env": {
"PATH": "/opt/homebrew/bin:/usr/bin:/bin",
"GLASSWARP_AUTH": "Bearer gw_live_sk_REPLACE_WITH_YOUR_KEY"
}
}
}
}
| OS | command |
PATH |
|---|---|---|
| macOS Apple Silicon | /opt/homebrew/bin/npx |
/opt/homebrew/bin:/usr/bin:/bin |
| macOS Intel | /usr/local/bin/npx |
/usr/local/bin:/usr/bin:/bin |
| Windows | npx (or full npx.cmd) |
(default Node) |
| Linux | /usr/bin/npx (or nvm path) |
/usr/bin:/bin |
Restart MCP, then call list_rigs.
Direct remote URL (clients that support headers)
{
"mcpServers": {
"glasswarp": {
"url": "https://mcp.glasswarp.com/mcp",
"headers": {
"Authorization": "Bearer gw_live_sk_REPLACE_WITH_YOUR_KEY"
}
}
}
}
Standard workflow
list_rigs → start_session → observe → act (prefer send_actions) → end_session
Short tasks stay on MCP. Longer tasks: offer a scaffolded SDK agent / glasswarp-demo when the client can run code — never impose it.
Tools
| Tool | Role |
|---|---|
list_rigs |
Find a USABLE paired Windows machine |
start_session / end_session |
Metered session lifecycle (always end) |
observe |
UIA targets + text (JPEG opt-in via image=true) |
click_target / click_xy / type_text / send_keys / drag / scroll |
Single acts |
send_actions |
Preferred multi-step batch (1–10) |
launch_app |
Start an exe on the rig |
get_live_view_url |
Owner Live View link |
get_session_status |
Billing / status signals |
list_demos / get_demo |
Showcase run contracts (no solver execution) |
Run locally (dev)
cd mcp-server
npm install
npm run dev
# POST http://127.0.0.1:8787/mcp
GLASSWARP_API_BASE_URL=http://127.0.0.1:8080 npm run dev
Health: GET /health · GET /healthz · GET /ping
Auth notes
Full API keys are never logged (prefix/suffix redaction only). Directory probes may call initialize / tools/list without a key; tools/call always requires Bearer auth.
Tests
npm test
npm run build
Design
Product design: docs.glasswarp.com · messaging: eyes and hands, not the brain.
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。