appcrane-mcp
MCP connector for AppCrane that serves a bundled tool catalog for offline introspection and proxies tool calls to a self-hosted AppCrane instance for deploying and managing AI-built apps.
README
appcrane-mcp
The standalone MCP connector for AppCrane — the self-hosted deployment platform for AI-built, agent-deployed apps.
AppCrane exposes an MCP server at <instance>/api/mcp that requires an X-API-Key. This package is a thin stdio MCP server that:
- Serves the full
appcrane_*tool catalog fortools/listintrospection with zero configuration — it ships a bundledcatalog.json, so registries and sandboxes (e.g. Glama) can start it in a container and introspect it offline. - Proxies real
tools/callto your own AppCrane instance onceAPPCRANE_URLandAPPCRANE_KEYare set.
It runs anywhere Node 20+ is available, via npx appcrane-mcp.
Install / run
npx appcrane-mcp
With no environment variables it still starts and answers tools/list from the bundled catalog (~35 tools). To actually invoke tools against your instance, set:
| Env var | Required | Description |
|---|---|---|
APPCRANE_URL |
for calls | Your instance base URL, e.g. https://crane.example.com |
APPCRANE_KEY |
for calls | Your AppCrane API key (sent as X-API-Key) |
APPCRANE_GITHUB_TOKEN |
optional | A GitHub PAT (sent as X-Github-Token) to unlock the github_* passthrough tools |
MCP client configuration
This is a stdio command server (not an HTTP endpoint). Add it to your MCP client config like this:
{
"mcpServers": {
"appcrane": {
"command": "npx",
"args": ["-y", "appcrane-mcp"],
"env": {
"APPCRANE_URL": "https://crane.example.com",
"APPCRANE_KEY": "your_api_key",
"APPCRANE_GITHUB_TOKEN": "ghp_optional_for_github_passthrough"
}
}
}
}
Claude Code CLI:
claude mcp add appcrane \
--env APPCRANE_URL=https://crane.example.com \
--env APPCRANE_KEY=your_api_key \
-- npx -y appcrane-mcp
How it works
tools/list- If
APPCRANE_URL+APPCRANE_KEYare set → connects to${APPCRANE_URL}/api/mcp(StreamableHTTP, headerX-API-Key, plusX-Github-Tokenwhen the GitHub token is set), fetches the live, auth-filtered tool list, and returns it. If the instance is unreachable it falls back to the bundled catalog. - If not configured → returns the bundled static catalog (
catalog.json). This path needs no env vars, which is what a registry's introspection sandbox hits.
- If
tools/call→ requiresAPPCRANE_URL+APPCRANE_KEY; forwards the call to the backend and returns its result. Without configuration it returns a clear "not configured" error.
Docker
docker build -t appcrane-mcp .
# Introspection only (no config) — starts and serves tools/list:
docker run -i --rm appcrane-mcp
# Full use:
docker run -i --rm \
-e APPCRANE_URL=https://crane.example.com \
-e APPCRANE_KEY=your_api_key \
appcrane-mcp
Development
npm install
npm run build # tsc -> dist/
npm run gen:catalog # regenerate catalog.json from the AppCrane platform source
node scripts/test-list.mjs # smoke test: spawn over stdio, assert tools/list
The bundled catalog.json is generated from the real tool definitions in the AppCrane platform (deployhub/server/services/mcpTools.js) and reflects the AWS-aligned tool vocabulary (appcrane_set_secret, appcrane_get_secret, appcrane_cp, and the stage parameter). It is committed so the package ships self-contained.
About AppCrane
AppCrane is a self-hosted home for AI-built and AI-deployed apps. This connector is MIT-licensed to maximize adoption; the platform itself is AGPL.
- Platform: https://github.com/gitayg/appCrane
- Product page: https://glick.run/appcrane.html
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。