procurement-mailer-mcp

procurement-mailer-mcp

Enables an SAP Joule agent to read and send email from a scoped Microsoft Exchange mailbox via MCP tools, translating tool calls into Microsoft Graph requests.

Category
访问服务器

README

procurement-mailer-mcp

A Model Context Protocol server that lets an SAP Joule agent read and send email from a scoped Microsoft Exchange mailbox. The agent never talks to Microsoft directly — this server sits in the middle, translating MCP tool calls into Microsoft Graph requests against a single mailbox.

📖 Setup runbook: https://fairconsult-corp.github.io/procurement-mailer-mcp/ 🏛️ Architecture: https://fairconsult-corp.github.io/procurement-mailer-mcp/architecture.html

Architecture

  SAP Joule Agent  ──▶  procurement-mailer-mcp  ──▶  Exchange / MS Graph
   (BTP / Joule)         (this repo, on CF)          (single scoped mailbox)

Invocation flows left-to-right; data flows back right-to-left. Three stages:

Stage What it does
1 · Joule Plans the request, picks an MCP tool, calls the server over Streamable HTTP through a BTP destination.
2 · MCP (this repo) Node.js + TypeScript server on Cloud Foundry. Exposes four tools and maps them onto Graph calls through the graph-mail destination (an Entra app identity).
3 · Graph Microsoft Graph over Exchange Online, restricted by an application access policy to just the demo mailbox.

Transport

The server is stateless: every POST /mcp builds a fresh McpServer + StreamableHTTPServerTransport pair (sessionIdGenerator: undefined). GET and DELETE on /mcp — the session channels of Streamable HTTP — return 405.

Graph access

There is no MSAL or token juggling in code. @sap-cloud-sdk/http-client's executeHttpRequest resolves the graph-mail BTP destination, performs the OAuth client-credentials exchange, and prefixes /v1.0 on every call. The Entra credentials (tenant ID, client ID, client secret) live in that destination, not in environment variables.

Project layout

src/
├── index.ts          # Express app + Streamable HTTP transport (POST /mcp)
├── server.ts         # Builds the McpServer, registers `ping`, wires in mail tools
├── tools/
│   └── mail.ts       # The three mail tools: list_messages, read_message, send_mail
└── graph/
    └── client.ts     # graphGet / graphPost via the `graph-mail` destination

manifest.yml          # Cloud Foundry app + service bindings (dest, conn, xsuaa)
index.html            # The setup runbook (published to GitHub Pages)
architecture.html     # Architecture page — call flow, components, data model
assets/               # Logo + runbook images

Tools

Tool Description Key inputs
ping Health check — returns pong @ <timestamp>. note?
list_messages Recent emails from the mailbox: id, subject, sender, received time, read state. top (1–25, default 10), unreadOnly
read_message Full body of one email by id. messageId
send_mail Send an email from the mailbox. to, subject, body, contentType (Text/HTML)

Requirements

  • Node.js 22.x (see engines in package.json)
  • For deployment: a SAP BTP subaccount with Cloud Foundry, and the graph-mail destination configured (Entra credentials). See the runbook for the full Microsoft-side and BTP setup.

Run locally

npm install

# Dev — tsx watch, hot reload, port 3000 (or $PORT)
npm run dev

# Or build to dist/ and run the compiled server
npm run build
npm start

The server listens on http://localhost:3000/mcp.

Validate with the MCP Inspector

Use a real MCP client — not curl, which can't complete the Streamable HTTP handshake.

npx @modelcontextprotocol/inspector
# Transport: Streamable HTTP   URL: http://localhost:3000/mcp
# Connect → List Tools (expect 4) → call `ping` → expect "pong @ <timestamp>"

ping works locally with no extra setup. The three mail tools resolve the graph-mail destination, which only exists on BTP — test those against the deployed route, or feed the SAP Cloud SDK a destinations env var when starting the server locally.

Deploy to Cloud Foundry

The app expects three bound services (see manifest.yml): a destination, connectivity, and xsuaa instance.

cf api <CF_API_ENDPOINT>          # e.g. https://api.cf.us10-001.hana.ondemand.com
cf login --sso
cf target -o "<ORG>" -s <SPACE>

# One-time: create the backing services manifest.yml binds
cf create-service destination lite  procurement-mailer-dest
cf create-service connectivity lite procurement-mailer-conn
cf create-service xsuaa application procurement-mailer-uaa

npm run build
cf push

# After re-binding or changing a service: RESTAGE, never restart
cf restage procurement-mailer-mcp

Restage, never restart. Service bindings are only injected into VCAP_SERVICES on cf restage. cf restart reuses the old droplet and the env stays empty.

Configuration

Setting Where Notes
MAILBOX manifest.yml env / src/graph/client.ts default The mailbox every call targets.
PORT injected by Cloud Foundry Defaults to 3000 locally.
Entra credentials graph-mail BTP destination Tenant ID, client ID, client secret — not env vars.

License

ISC

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