moolre-mcp
MCP server that exposes Moolre's API as 24 tools for AI agents, enabling account management, transfers, payments, SMS, and WhatsApp operations.
README
moolre-mcp (Node.js)
MCP server exposing Moolre's API — accounts, transfers, payments, SMS, WhatsApp — as 24 tools an AI agent can call. This is the Node.js port of the original PHP implementation (mcp_php/mcp.php); behavior is a 1:1 match, verified against it directly.
Built on the official @modelcontextprotocol/sdk for protocol handling (JSON-RPC validation, session management, SSE framing) — Moolre-specific logic (the REST client and 24 tool definitions) is the only custom code.
Requirements
- Node.js 18 or later (uses the built-in
fetch; developed and tested on Node 22). npm install(pulls in@modelcontextprotocol/sdk).
Running locally (stdio — for MCP clients like Claude Code, Cursor)
node mcp.js
When run with a plain node mcp.js and no PORT env var, this reads/writes newline-delimited JSON-RPC over stdin/stdout — the shape an MCP client expects when it spawns the server as a subprocess, e.g. "command": "node", "args": ["/path/to/mcp.js"] in the client's MCP config. See CURSOR_SETUP.md / AGENT_SETUP.md for full config examples.
Running as a hosted HTTP server
PORT=3000 node mcp.js
Setting PORT (which most Node hosts — Render, Railway, Fly.io, etc. — inject automatically) switches this to the Streamable HTTP transport: a single endpoint handling
POST— JSON-RPC messages (responds as plain JSON, or as an SSE-framed event if the client sendsAccept: text/event-stream)GET— optional standalone SSE stream (used for server-initiated messages; this server has none to push, so it opens and immediately closes the stream — enough to satisfy clients that check for SSE support)DELETE— explicit session termination
Session IDs are issued on initialize via the Mcp-Session-Id response header (handled by the SDK's StreamableHTTPServerTransport), and required on every subsequent request via the same header. We keep one transport (and one connected MCP Server instance) per session in an in-memory Map — if you ever run multiple instances behind a load balancer without sticky sessions, session lookups will fail against the wrong instance. For a single hosted instance this isn't a concern.
To force HTTP mode without a PORT (rare), set MCP_TRANSPORT=http instead (defaults to port 3000).
Environment variables
| Variable | Purpose | Used by |
|---|---|---|
X-API-USER |
Your Moolre username | every endpoint |
X-API-KEY |
Private API key | account/transfer/payment/list_transactions endpoints |
X-API-PUBKEY |
Public API key | create_payment_id, create_bank_account_number, generate_payment_link, payment_status |
X-API-VASKEY-SMS |
VASKEY for SMS endpoints | send_sms, send_sms_get, sms_status, create_sender_id, sender_id_status, list_sender_ids, approve_sender_id, sms_account_status |
X-API-VASKEY-WHATSAPP |
VASKEY for WhatsApp endpoints | get_whatsapp_templates, send_whatsapp_message, whatsapp_message_status |
accountnumber |
Your Moolre wallet account number | most endpoints (not create_account, and not SMS/WhatsApp endpoints) |
All of these are fallbacks only — if a tool call explicitly supplies headers or accountnumber in its arguments, that always takes precedence over the environment.
The single header Moolre actually receives is always X-API-VASKEY — the two env vars just let the server pick the right value automatically depending on which service (SMS vs WhatsApp) the tool being called talks to.
Project structure
mcp.js— REST client (callRest), the MCPServerwiring (tools/list/tools/callhandlers), and both transports (stdio viaStdioServerTransport, HTTP viaStreamableHTTPServerTransport+ a thin session-routing layer)tools.js— the 24 tool schemas and their REST-call configuration (method, path, defaulttype, credential group)
Testing
# stdio
printf '%s\n%s\n' \
'{"jsonrpc":"2.0","id":1,"method":"initialize","params":{"protocolVersion":"2025-06-18","capabilities":{},"clientInfo":{"name":"test","version":"1.0"}}}' \
'{"jsonrpc":"2.0","id":2,"method":"tools/list","params":{}}' \
| node mcp.js
# HTTP
PORT=3000 node mcp.js &
curl -s -i -X POST http://127.0.0.1:3000/ \
-H 'Content-Type: application/json' -H 'Accept: application/json, text/event-stream' \
-d '{"jsonrpc":"2.0","id":1,"method":"initialize","params":{"protocolVersion":"2025-06-18","capabilities":{},"clientInfo":{"name":"test","version":"1.0"}}}'
Note: the SDK validates requests against the real MCP schema, so initialize needs a proper clientInfo/capabilities payload (real MCP clients already send this) — unlike the earlier hand-rolled version, bare {"params":{}} won't pass validation.
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。