serverless-mcp-server logo

serverless-mcp-server

by eleva

A simple Model Context Protocol (MCP) server deployed on AWS Lambda and exposed via Amazon API Gateway, deployed with Serverless Framework. It provides a minimal setup for interacting with models using the MCP standard.

View on GitHub

Last updated: N/A

🧠 serverless-mcp-server

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 FrΓ©dΓ©ric Barthelet: which has developed a middy middleware for Model Context Protocol (MCP) server integration with AWS Lambda functions in this repo

πŸ›  Features

  • πŸͺ„ Minimal MCP server setup using @modelcontextprotocol/sdk
  • πŸš€ Deployed as a single AWS Lambda function
  • 🌐 HTTP POST endpoint exposed via API Gateway at /mcp
  • πŸ”„ Supports local development via serverless-offline
  • πŸ§ͺ Includes a simple example tool (add) with JSON-RPC interaction

πŸ“¦ Project Structure

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

πŸ›  Prerequisites

πŸš€ Getting Started

  1. Install dependencies:
npm install
  1. Install open source severless globally (if not already installed):
npm install -g osls
  1. 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.

Switch to Api Gateway V2 (HTTP API)

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.

πŸ§ͺ Test with curl requests

List tools

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
}'

βž• Use the add Tool

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
    }
  }
}'

🧬 Code Breakdown

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

src/index.js

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());

πŸ“‘ Deploy to AWS

Just run:

sls deploy

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

πŸ“˜ License

MIT β€” feel free to fork, tweak, and deploy your own version!