rpmc-superops-mcp
Provides authenticated, read-only access to SuperOps data via MCP, supporting stdio and HTTP transports with privacy-safe redaction.
README
rpmc-superops-mcp
Standalone SuperOps MCP server for RPM Computers (RPMC).
This is not a fork of WYRE, Computask, or Servosity. Those projects are implementation donors and live-API references. See docs/PROVENANCE.md.
Phase 1: authenticated, read-only, Docker on QNAP (LAN). Write tools are not registered and do not appear in tools/list. Constrained search and investigation tools are documented in docs/READ-SURFACE.md. Complete official get* accounting is in docs/OFFICIAL-READ-INVENTORY.md. The next staging pass should follow docs/LIVE-CONFIRMATION-MATRIX.md. After a QNAP image with a new tool surface, fully reconnect the Cursor MCP client.
Runtime
- Node.js 24
- MCP TypeScript SDK v2 (
@modelcontextprotocol/server+@modelcontextprotocol/node) - stdio (desktop MCP clients) or Streamable HTTP (QNAP)
- SuperOps credentials only from container/process environment
- HTTP MCP callers must send
Authorization: Bearer <MCP_AUTH_TOKEN>(MCP_AUTH_TOKEN≥ 32 characters) - stdio does not require
MCP_AUTH_TOKEN rpmc_statusreportscommitfrom image envRPM_BUILD_COMMIT(Docker build-argGIT_COMMIT; local/dev fallbackunknown). No Git at runtime.
See docs/MCP-SDK.md for Host/Origin policy and the v2 decision.
Quick start (development)
cp .env.example .env
npm install
npm test
npm run build
HTTP:
MCP_TRANSPORT=http MCP_AUTH_TOKEN=... SUPEROPS_API_TOKEN=... SUPEROPS_SUBDOMAIN=... SUPEROPS_REGION=us npm start
Callers:
POST /mcp
Authorization: Bearer <MCP_AUTH_TOKEN>
Accept: application/json, text/event-stream
HTTP is stateless (fresh MCP server per request) so it works behind a future Cloudflare Access/Tunnel hop without sticky sessions. /health is unauthenticated for Docker HEALTHCHECK and does not expose tokens or ticket content.
If a browser client will send Origin, set MCP_ALLOWED_ORIGINS. For LAN QNAP or a future tunnel hostname, set MCP_ALLOWED_HOSTS (or set Origins and let Host reuse that list). Non-browser clients that omit Origin continue to work; a present Host is always required on /mcp.
Docker
docker build -t rpmc-superops-mcp:local .
docker run --rm -p 127.0.0.1:8080:8080 --env-file /secure/path/.env rpmc-superops-mcp:local
See docker-compose.sample.yml. Production compose stays on QNAP and is not committed.
CI builds the image on every push (no registry publish). Local/QNAP image smoke-test is still required on a machine with Docker.
Privacy
Tool JSON payloads run through a conservative safe-output pass: high-confidence secrets in freeform strings are replaced with [redacted] and marked via _privacy when anything changed. Conversation/note content is also HTML-stripped. Attachments stay metadata-only. Useful technical evidence is kept. This is not DLP.
Freeform ticket/alert bodies are not general-purpose email redaction. Emails in DESCRIPTION/conversation/note/alert text may be technically relevant and are left in place. Aggregators omit structured requester.email (and similar structured email keys on user objects) while keeping id/name.
Stderr audit logs (mcp.tool_call) record tool name, success, outcome (complete | partial | failed), section state, resolution method, truncation, duration, and safe upstream failure category. They must not contain ticket bodies, subjects, names, emails, IPs, tokens, or raw SuperOps responses. A failed or partial investigation is success: false with an explicit outcome, even when the tool returns structured JSON.
License
Apache-2.0. See LICENSE, NOTICE, and THIRD_PARTY_NOTICES.md.
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。