IFTTT MCP Proxy
Enables stdio-only MCP clients to connect to IFTTT's remote MCP server by handling OAuth authentication, SSE streaming, and response transformation.
README
IFTTT MCP Proxy
A local stdio MCP server that bridges IFTTT's remote MCP endpoint to any MCP client that only supports the stdio transport (e.g., Amazon Quick, Claude Desktop, Cursor, etc.).
The Problem
IFTTT exposes a remote MCP server at https://ifttt.com/mcp using:
- OAuth 2.1 with PKCE for authentication
- Streamable HTTP transport with SSE (Server-Sent Events) for streaming responses
Many MCP clients (including Amazon Quick and Claude Desktop) only support stdio transport — they launch a local process and communicate via stdin/stdout. They cannot directly connect to remote HTTP MCP servers that require OAuth redirect flows.
The Solution
This proxy runs as a local Node.js process that:
- Handles OAuth 2.1 + PKCE — one-time browser-based authentication flow with automatic token refresh
- Bridges stdio ↔ HTTP — reads JSON-RPC messages from stdin, forwards them to IFTTT's remote endpoint, and writes responses to stdout
- Handles SSE streaming — IFTTT responds to tool calls with HTTP 202 + SSE streams; the proxy keeps the connection open and collects the full response before writing it to stdout
- Transforms responses — IFTTT returns data in
structuredContentwhile standard MCP clients expect it incontent[]; the proxy normalizes this automatically - Auto-refreshes tokens — detects expired tokens and uses the refresh token to obtain new access tokens transparently
Architecture
┌─────────────┐ stdio ┌──────────────────┐ HTTPS/SSE ┌─────────────┐
│ MCP Client │ ──stdin/stdout──│ ifttt-mcp-proxy │ ───────────────│ ifttt.com/ │
│ (e.g. │ │ │ │ mcp │
│ Amazon │ │ • OAuth token │ │ │
│ Quick) │ │ • SSE handling │ │ (Remote │
│ │ │ • Response │ │ MCP │
│ │ │ transform │ │ Server) │
└─────────────┘ └──────────────────┘ └─────────────┘
Installation
git clone https://github.com/maishsk/ifttt-mcp-proxy.git
cd ifttt-mcp-proxy
No dependencies required — uses only Node.js built-in modules (http, https, crypto, fs, path, url, child_process).
Requirements: Node.js 18+
Usage
1. Authenticate (one-time)
node index.js --auth
This opens your browser for the IFTTT OAuth flow. After authorizing, the token is saved to ~/.quickwork/ifttt-token.json (with 0600 permissions).
2. Check Token Status
node index.js --status
Shows whether your token is valid, when it was obtained, and when it expires.
3. Run as MCP Proxy
node index.js
Starts in stdio mode — reads JSON-RPC from stdin, proxies to IFTTT, writes responses to stdout. Diagnostic logs go to stderr.
4. Configure in Your MCP Client
Add to your MCP client configuration:
{
"mcpServers": {
"ifttt": {
"command": "node",
"args": ["/path/to/ifttt-mcp-proxy/index.js"]
}
}
}
Or use the included wrapper script that handles first-run auth automatically:
{
"mcpServers": {
"ifttt": {
"command": "bash",
"args": ["/path/to/ifttt-mcp-proxy/start.sh"]
}
}
}
How It Works
OAuth 2.1 + PKCE Flow
- Generates a PKCE code verifier/challenge pair
- Opens browser to
https://ifttt.com/oauth/authorizewith the challenge - Spins up a temporary HTTP server on
localhost:3118to receive the callback - Exchanges the authorization code for an access token + refresh token
- Saves tokens to
~/.quickwork/ifttt-token.json
Streamable HTTP + SSE Handling
IFTTT's MCP server uses the Streamable HTTP MCP transport:
- Simple requests (like
initialize,tools/list) return direct JSON responses - Tool calls (
tools/call) may return HTTP 200/202 withContent-Type: text/event-stream - The SSE stream contains
data:lines with the actual JSON-RPC response
The proxy handles all three response types:
- Direct JSON (200 +
application/json) → parsed and forwarded - SSE stream (200/202 +
text/event-stream) → collected until stream ends, events parsed and forwarded - Error responses (4xx/5xx) → converted to JSON-RPC error objects
Response Transformation
IFTTT puts tool results in a non-standard structuredContent field while leaving the standard content array empty. The proxy detects this and moves the data into content[{type: "text", text: ...}] for compatibility.
Token Auto-Refresh
On every request, the proxy checks token expiry (with a 60-second buffer). If expired, it uses the refresh token to obtain a new access token before proceeding. If a request returns HTTP 401, it attempts a refresh and retries once.
Files
| File | Description |
|---|---|
index.js |
Main proxy server (OAuth + stdio proxy) |
package.json |
Package metadata |
start.sh |
Wrapper script with auto-auth on first run |
Token Storage
Tokens are stored at ~/.quickwork/ifttt-token.json with file permissions 0600 (owner read/write only). The file contains:
access_token— Bearer token for IFTTT APIrefresh_token— Used to obtain new access tokensexpires_in— Token lifetime in secondsobtained_at— Timestamp when the token was obtained
Troubleshooting
"No token found"
Run node index.js --auth to complete the OAuth flow.
"Token refresh failed"
Your refresh token may have been revoked. Re-run node index.js --auth.
Empty responses from tool calls
This usually means the SSE stream wasn't properly consumed. Check stderr logs for details. The proxy handles the common IFTTT pattern of 202 + SSE.
Port 3118 in use
The OAuth callback uses port 3118. If it's occupied, kill the process using it or wait for it to release.
License
MIT
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。