MCP Server Starter

MCP Server Starter gives you a basic structure to run local tools with Cursor, Claude, and others using the MCP standard. It supports multiple transport methods, simple setup, and TypeScript typings to help you build and test MCP-based workflows locally or remotely.

How to use

You connect an MCP client to your server to execute tools and manage workflows. You can run locally using standard I/O, or expose the server over HTTP so remote clients can send requests. For local development and testing, start the server in stdio mode or use an HTTP endpoint for remote usage. The Inspector helps you debug and tune your server while it runs.

How to install

Prerequisites you need on your machine before starting are Node.js and a compatible MCP client such as Cursor. Ensure Node.js is installed and available in your system path.

Step 1. Install dependencies for your MCP server (if your project includes a package manager setup).

Step 2. Run in local development via stdio or start an HTTP server for remote usage.

Example configurations

{
  "mcpServers": {
    "starter_http": {
      "url": "http://localhost:4200/mcp"
    },
    "starter_stdio_node": {
      "command": "node",
      "args": ["./bin/cli.mjs", "--stdio"]
    },
    "starter_stdio_npx": {
      "command": "npx",
      "args": ["my-mcp-server", "--stdio"]
    }
  }
}

Notes on transport options and usage

Supported transports let you choose between local development (stdio) and remote or mixed usage (HTTP). The stdio option runs locally via command line, while the HTTP option exposes a REST-like endpoint for remote clients.

The Streamable HTTP transport is available for remote and local use, while Server-Sent Events (sse) is deprecated but shown for compatibility in some examples.

Use the Inspector

To debug your MCP server, run the inspector to inspect requests, responses, and tool execution paths.

Available tools

inspect

Use the inspect command to debug and inspect the MCP server during development and troubleshooting.