AItomation MCP relay

AItomation MCP relay

Enables external AI applications to access MCP tools by providing a user-hosted HTTPS endpoint that forwards requests to a paired desktop client.

Category
访问服务器

README

AItomation MCP relay

User-hosted HTTPS edge for inbound AI connectors. Other apps call this process on an address you own. The app on your computer pairs to it later and runs the tools. This process never runs tools itself.

This repository is the connector process only. It is not the desktop app and it is not the Team Space team server.

We do not host this door for you. Place it on a server you control. A desktop app update does not update this process. A desktop v* tag does not publish this image.

One process, one project

One relay talks to one device and one project.

  • Three projects that need a public HTTPS door need three processes.
  • Path prefixes on one name (/project-a/mcp) are refused.
  • The Team Space server on the same machine is a fourth process. Same machine is fine. Same process is not.

Quick start (Docker)

docker pull ghcr.io/stanislavmandrik621/aitomation-mcp-relay:latest
docker run --rm -p 8790:8790 \
  -e MCP_RELAY_HOST=0.0.0.0 \
  -e MCP_RELAY_PORT=8790 \
  -e MCP_RELAY_DATA_DIR=/data \
  -v mcp-relay-data:/data \
  ghcr.io/stanislavmandrik621/aitomation-mcp-relay:latest

Default listen port is 8790. Do not use 8787, 8788, or 8789.

GET / and GET /health answer 200. A browser shows a short status page. curl and a later pair probe (Accept: text/plain) still get the one-line text AItomation MCP relay.

POST /mcp without a sign-in token returns 401 plus a WWW-Authenticate header that points at the well-known JSON. After you pair the app and complete OAuth, that POST is forwarded to the app. Tools never run on this process. Pairing uses /pair/hello, /pair/next, /pair/reply, and /pair/unavailable.

Official images are multi-arch (linux/amd64 and linux/arm64). Prefer pinning a version tag in production.

Public certificate on a public host

Cloud hosts that call this connector (ChatGPT, Claude, Grok, Gemini Enterprise) need a certificate from a public certificate authority. Do not use a self-made certificate on a public host. Put a reverse proxy with a public certificate in front of this process, or pass MCP_RELAY_TLS_CERT_FILE and MCP_RELAY_TLS_KEY_FILE that already come from that authority.

Environment

Name Default Notes
MCP_RELAY_HOST 127.0.0.1 Docker sets 0.0.0.0.
MCP_RELAY_PORT 8790 Reserved ports are refused.
MCP_RELAY_DATA_DIR ./data Use an absolute path under a service manager.
MCP_RELAY_PAIR_SECRET unset Pair secret stays in env, never in argv.
MCP_RELAY_PUBLIC_URL unset Public https origin (optional /mcp). Used in well-known resource.
MCP_RELAY_STATIC_HEADER_HASH unset Optional SHA-256 hex of the Claude request-header key for the paired project.
MCP_RELAY_TLS_CERT_FILE unset Optional. Public CA for a public host.
MCP_RELAY_TLS_KEY_FILE unset Pair with the cert file.

This process does not read Team Space environment names.

OAuth stays on port 8790. Authorization URL is /authorize. Token URL is /token. Include offline_access in scopes. Gemini Enterprise uses the confidential client id minted on first start (secret is written once under the data dir) and redirect https://vertexaisearch.cloud.google.com/oauth-redirect. Well-known JSON is at /.well-known/oauth-protected-resource and /.well-known/oauth-protected-resource/mcp. There is no HTML catch-all on those paths.

Alternate (clone + npm)

git clone https://github.com/stanislavmandrik621/aitomation-mcp-relay.git
cd aitomation-mcp-relay
npm ci
npm start

npm start is for development. For a real service, build once (npm run build) and run node dist/server.js under a service manager.

Step-by-step service unit, three-project compose, and HTTPS proxy snippets: docs/SELF-HOST.md.

Stop

SIGTERM or SIGINT stops new work, waits for in-flight POST /mcp forwards to finish, then exits. The process does not kill itself mid-POST.

Source of truth

Day-to-day edits live in the AItomation monorepo packages/mcp-relay and are published here for customers. Official GHCR write is this repo's publish-image workflow. The desktop cut does not publish this image.

License

MIT - see LICENSE.

推荐服务器

Baidu Map

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。

官方
精选
JavaScript
Playwright MCP Server

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。

官方
精选
TypeScript
Magic Component Platform (MCP)

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。

官方
精选
本地
TypeScript
Audiense Insights MCP Server

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。

官方
精选
本地
TypeScript
VeyraX

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。

官方
精选
本地
graphlit-mcp-server

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。

官方
精选
TypeScript
Kagi MCP Server

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。

官方
精选
Python
e2b-mcp-server

e2b-mcp-server

使用 MCP 通过 e2b 运行代码。

官方
精选
Neon MCP Server

Neon MCP Server

用于与 Neon 管理 API 和数据库交互的 MCP 服务器

官方
精选
Exa MCP Server

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。

官方
精选