MCP Server Deployment Guide

---
title: "MCP Server Deployment Guide"
description: "Guide non-developer users through deploying MCP servers on Cloudflare Workers (free), Railway ($5+), or Docker (self-hosted). Use when asked to help deploy, host, make available remotely, or publish an MCP server so an agent runtime can connect to it."
author: AgentArmory
license: Proprietary
---

# MCP Server Deployment Guide

Guide non-developer users through deploying MCP servers on Cloudflare Workers (free), Railway ($5+), or Docker (self-hosted). Use when asked to help deploy, host, make available remotely, or publish an MCP server so an agent runtime can connect to it.

## When to Use

Use when a user asks for help deploying, publishing, or hosting an MCP server. The user may not know they need deployment help -- they may say "my MCP server only works on my laptop" or "I want my agent to use this MCP server from anywhere" or "can you set this up so I don't need to run a terminal every time."

This skill is for **non-developer users**: small business owners, solopreneurs, freelancers, and personal-assistant users who run agent applications (OpenClaw, Hermes, Claude Desktop) and need their MCP servers accessible remotely without learning infrastructure tooling.

## Triggers

"deploy my MCP server", "host my skill", "my agent needs API access from anywhere", "make my MCP server available remotely", "Cloudflare Workers MCP", "Railway deploy MCP", "Docker MCP server", "run MCP server in the cloud", "remotely hosted MCP", "set up my MCP server for production", "I want my agent to use this server from my phone too"

## Prerequisites

Before deploying, the user must have:
- An existing MCP server project (Python FastMCP, Node.js MCP SDK, or similar) that works locally via stdio
- A GitHub account (free) for auto-deploy workflows on Cloudflare and Railway
- For self-hosted Docker: access to a server or computer that stays running (home server, VPS, or cloud VM)

## Methodology

### Phase 0: Deployability Assessment

Before choosing a target, run this checklist to determine which deployment option fits:

| Question | If yes | If no |
|----------|--------|-------|
| Does the server need to run for free? | Cloudflare Workers (free tier) | Railway or Docker |
| Does the server have complex filesystem or database dependencies? | Docker | Cloudflare or Railway |
| Does the user already have a Cloudflare account? | Consider Workers first | Consider Railway |
| Does the user need maximum control and zero vendor lock-in? | Docker (self-hosted) | Cloudflare or Railway |
| Does the server use SSE transport or need long-lived connections? | Railway or Docker (Workers has 30s CPU timeout) | Cloudflare Workers for simple request/response |
| Is the server a Node.js or Python application? | Any option | Containerized first for other runtimes |

### Phase 1: Prepare the Server for Deployment

Before any deployment, the MCP server must be modified for remote access:

1. **Change transport from stdio to SSE or Streamable HTTP.** The local development transport (`stdio`) connects the agent and server as child processes on the same machine. For remote deployment, the server must listen on an HTTP endpoint using Server-Sent Events (SSE) or Streamable HTTP transport.

   For Python FastMCP: