How to create an MCP Server

Faik Uygur

5 min read
How to create an MCP Server


Getting Started with Model Context Protocol (MCP)

The Model Context Protocol (MCP) is an open standard that makes it easy for applications to work with large language models (LLMs). Think of MCP as a “USB-C port for AI” a simple, unified way for various data sources, tools, and services to plug into your LLMs without needing custom code for every connection.

How is MCP Used?

  • Building Smart Agents and Workflows:
    MCP lets you combine different data streams and services into one smart AI agent. This means your LLM gets richer, more useful context without extra hassle.
  • Standardized Integrations:
    With a growing ecosystem of ready-to-use connectors, your LLM can seamlessly interact with files, databases, APIs, and more, no custom integrations required. You can now explore and discover MCP server directories here to get ideas of what is possible:
    https://cursor.directory/
    https://github.com/punkpeye/awesome-mcp-servers
    https://mcp.so/
  • Vendor Flexibility:
    MCP reduces the risk of being tied to a single vendor, letting you switch between LLM providers effortlessly based on your needs.

Core Building Blocks of MCP

MCP is built on three key components:

  • Tools:
    These are functions that let LLMs do things like run commands, call APIs, or perform calculations automatically. In Robomotion you can turn your automation flows into tools to your LLM.
  • Resources:
    This is the data your server shares, such as text, logs, images, or live data, to help LLMs produce accurate responses. You can use Robomotion to query databases, read from filesystem or do http requests to provide the required external resources to your LLM.
  • Prompts:
    Reusable templates that guide how your LLM interacts with users. They help ensure that every conversation follows a clear, predictable format.

Together, these components help you build AI applications that are both powerful and secure.

Communication Options for Your MCP Server

When you set up an MCP server, you have two main ways to handle communication between your server and clients:

1. STDIO Transport

  • How It Works:
    The server communicates via the console using the JSON-RPC protocol, making it ideal for local setups. Most MCP servers are built as command-line tools that read from and write to the console. However, the downside is that your application cannot output anything else to the console doing so would break the protocol between the client (host) and the MCP server.

Using the MCP Inspector with STDIO:
Start the MCP Inspector with:

npx @modelcontextprotocol/inspector your_mcp_server.py arg1 arg2

Replace your_mcp_server.py with your server script and add any extra arguments needed.

2. HTTP SSE Transport

  • How It Works:
    This method uses HTTP Server-Sent Events (SSE) for real-time communication, ideal for remote or cloud deployments.

    Robomotion RPA supports HTTP SSE, offering a scalable and easy-to-configure solution.

Using the MCP Inspector with HTTP SSE:
Simply run the command below, then open your browser and go to http://127.0.0.1:6274.

npx @modelcontextprotocol/inspector
Note: Make sure Node.js is installed to use the npx command. You can download it from Node.js official website.

Accelerating Your Development with Robomotion RPA

Robomotion makes it simple to build your MCP server with a visual, drag-and-drop interface. Here’s how:

Create a New Project:
Open Robomotion, create a new project, and install the MCP package from the packages section.

Set Up Your Listener:
Drag the Listen node onto your workspace and configure it with your desired IP address, port, and endpoint for HTTP SSE. This node has special outputs for tools, prompts, and resources.

Add a Calculator Tool:

  • Drag a Tool In node and connect it to a Function (Simple Add) node.
  • Rename the function to “Simple Add.”
  • Configure the Tool In node by adding a description like “Simple tool to add two numbers” and set the tool name to “add_numbers.”


Implement the Functionality:
Double-click the “Simple Add” function node and insert this code:

msg.result = msg.parameters.a + msg.parameters.b;
return msg;

Define the JSON Schema:
Double-click the Tool In node to open the editor, then replace its content with the following JSON schema. For more details on this format, please visit OpenAI's Function Calling Guide.

This tells the LLM which parameters to pass to your tool:

{
  "type": "object",
  "properties": {
    "a": {
      "type": "number",
      "description": "The first number to add."
    },
    "b": {
      "type": "number",
      "description": "The second number to add."
    }
  },
  "required": ["a", "b"],
  "description": "Schema for adding two numbers using parameters a and b."
}

Run and Test Your MCP Server

Now that your MCP server is ready in Robomotion RPA, follow these final steps:

  1. Download and Connect the Robomotion Robot (if needed):
    If you haven’t done so, download the Robomotion Robot from Robomotion Downloads, install it, and connect it to your workspace. For additional help, you can watch this YouTube tutorial. You must have at least v25.4.2 to run this automation.
  2. Run the Flow Now:
    Run your flow within Robomotion Designer to start your MCP server with the newly added tool.
  3. Test with MCP Inspector:
    Open the MCP Inspector to see your server in action. The inspector lets you view endpoints, run your tool, and check responses in real time.

Set Up Claude Desktop with Your MCP Server

Optionally, you can configure Claude Desktop to use your MCP server. After running your MCP server flow, use the following configuration:

{
  "mcpServers": {
    "robomotion": {
      "command": "npx",
      "args": ["mcp-remote", "http://127.0.0.1:8888/sse"]
    }
  }
}

This allows Claude Desktop to connect directly to your locally running MCP server.

Robomotion MCP support - Watch Video

🚀 Conclusion

The Model Context Protocol (MCP) streamlines the way your AI interacts with tools, data, and services, acting like a universal port for intelligence. With Robomotion RPA’s drag-and-drop builder and native MCP integration, you can turn an idea into a functional, context-aware AI agent in just minutes. No complex setup. No custom integrations.

Whether you're building smart assistants, automating workflows, or connecting APIs, MCP + Robomotion provides the flexibility and speed to deliver real value fast.

Here’s the flow shown in the video:

🔧 https://learn.robomotion.io/designer/shared/XfN4tz94HrNtzy6kLMqWnB

💬 Got questions? Join our community on Discord: community.robomotion.io

← Back to all posts