A super simple Model Context Protocol (MCP) server deployed on AWS Lambda and exposed via Amazon API Gateway, deployed with Serverless Framework. This skeleton is based on the awesome work of Frédéric Barthelet: which has developed a middy middleware for Model Context Protocol (MCP) server integration with AWS Lambda functions in this repo

📖Read the full article here on dev.to

• 🪄 Minimal MCP server setup using @modelcontextprotocol/sdk
• 🚀 Deployed as a single AWS Lambda function
• 🌐 HTTP POST endpoint exposed via API Gateway at /mcp
• 🔄 Supports local development via serverless-offline
• 🧪 Includes a simple example tool (add) with JSON-RPC interaction

serverless-mcp-server/
├── src/                    # Source code
│   └── index.js                # MCP server handler
├── .gitignore              # Git ignore file
├── package.json            # Project dependencies
├── package-lock.json       # Project lock file
├── README.md               # This documentation file
└── serverless.yml          # Serverless Framework config

• Node.js v22+
• Open Source Serverless or Serverless Framework v3+
• Serverless Offline

1. Install dependencies:

npm install

2. Install open source severless globally (if not already installed):

npm install -g osls

3. Run Locally with serverless-offline

npm sls offline

Local endpoint will be available at: POST http://localhost:3000/dev/mcp

Note that the /dev/ stage is added by default when using serverless-offline, reflecting Api Gateway V1 (REST API) behavior.

If you want to use API Gateway V2, you can change the serverless.yml file to use httpApi instead of http in the events section. This will allow you to use HTTP APIs instead of REST APIs. This will allow you to use HTTP APIs instead of REST APIs.

functions:
  mcpServer:
    handler: src/index.handler
    events:
      - httpApi:
          path: mcp
          method: post

Local endpoint will be available at: POST http://localhost:3000/mcp

Note that the /dev/ stage is not needed when using API Gateway V2. Note you should change test curl and postman requests accordingly.

curl --location 'http://localhost:3000/dev/mcp' \
--header 'content-type: application/json' \
--header 'accept: application/json' \
--header 'jsonrpc: 2.0' \
--data '{
  "jsonrpc": "2.0",
  "method": "tools/list",
  "id": 1
}'

curl --location 'http://localhost:3000/dev/mcp' \
--header 'content-type: application/json' \
--header 'accept: application/json' \
--header 'jsonrpc: 2.0' \
--data '{
  "jsonrpc": "2.0",
  "id": 2,
  "method": "tools/call",
  "params": {
    "name": "add",
    "arguments": {
      "a": 5,
      "b": 3
    }
  }
}'

There are some basic tests included in the __tests__ folder. You can run them with:

npm run test

This code is based on the awesome work of Frédéric Barthelet: which has developed a middy middleware for Model Context Protocol (MCP) server integration with AWS Lambda functions in this repo

import middy from "@middy/core";
import httpErrorHandler from "@middy/http-error-handler";
import { McpServer } from "@modelcontextprotocol/sdk/server/mcp.js";
import { z } from "zod";
import mcpMiddleware from "middy-mcp";

const server = new McpServer({
  name: "Lambda hosted MCP Server",
  version: "1.0.0",
});

server.tool("add", { a: z.number(), b: z.number() }, async ({ a, b }) => ({
  content: [{ type: "text", text: String(a + b) }],
}));

export const handler = middy()
  .use(mcpMiddleware({ server }))
  .use(httpErrorHandler());

Just run:

sls deploy

After deployment, the MCP server will be live at the URL output by the command.

This repository has been quoted in the following newsletters:

• Serverless Advocate Newsletter #33 by Lee Gilmore (AWS Hero)

MIT — feel free to fork, tweak, and deploy your own version!