MCP Server example using OpenAI from Marcelo

[marcelocecin]

I would like to request support for the MCP server role.

[kensoh]

Hi @marcelocecin can you tell me more what is needed here? Is it a request to add enhancement for MCP integration with chatbots?

[marcelocecin]

Perfect — let's build you a full-stack demo showing:

OpenAI (GPT-4-turbo or GPT-4o) calling an MCP server that runs Tebelorg RPA Python scripts in real time.

---

📦 Full-Stack Demo Plan

Here’s what we'll do:

Step | Part -- | -- 1 | Create an MCP server exposing RPA tools 2 | Use ngrok to make it public 3 | Write a Python OpenAI client that defines and calls the RPA tool 4 | See it all working

---

📄 1. Create the MCP server with RPA-Python

Save this as mcp_rpa_server.py:

# mcp_rpa_server.py
from mcp.server.fastmcp import FastMCP, tool

import rpa as r
@tool("automation://fill-web-form")

def fill_web_form(name: str, email: str) -> str:

r.init()

r.url("https://www.w3schools.com/html/html_forms.asp")  # public test form

r.type('//[@id="fname"]', name)

r.type('//[@id="lname"]', email.split('@')[0])  # fake last name

r.click('//*[@value="Submit"]')

r.close()

return f"Form submitted for {name} ({email})"
app = FastMCP()

app.add_tool(fill_web_form)
if name == "main":

app.run()

📌 What it does:

• Exposes fill_web_form as an MCP tool.

• Uses Tebelorg RPA to fill a form on a test site.

---

🌍 2. Expose the server to the internet

Install ngrok:

pip install pyngrok

Start ngrok tunnel:

ngrok http 3333

You will get something like:

Forwarding https://1234abcd.ngrok-free.app -> http://localhost:3333

✅ Copy your public ngrok URL!

---

💬 3. Create the OpenAI Client

Save this as openai_client.py:

# openai_client.py

import openai

import requests
openai.api_key = "your-openai-api-key-here"
MCP_SERVER_URL = "https://1234abcd.ngrok-free.app"  # replace with your ngrok URL
Define the tool (function)
functions = [

{

"name": "fill_web_form",

"description": "Fills out a web form with a name and email using RPA",

"parameters": {

"type": "object",

"properties": {

"name": {"type": "string", "description": "Person's name"},

"email": {"type": "string", "description": "Person's email address"},

},

"required": ["name", "email"]

}

}

]
Ask GPT to decide if it should call the function
response = openai.ChatCompletion.create(

model="gpt-4-turbo",  # or "gpt-4o"

messages=[

{"role": "system", "content": "You can automate web forms using a special tool."},

{"role": "user", "content": "Please fill a form with name Alice and email alice@example.com"}

],

functions=functions,

function_call="auto"  # Let GPT decide

)
Check if a function call is requested
if "function_call" in response["choices"][0]["message"]:

function_call = response["choices"][0]["message"]["function_call"]

print("[GPT Requested Call]:", function_call)
payload = {
    "name": function_call["arguments"]["name"],
    "email": function_call["arguments"]["email"],
}

# Send to our MCP server
print("[Calling MCP server...]")
mcp_response = requests.post(
    f"{MCP_SERVER_URL}/automation://fill-web-form",
    json=payload
)
print("[MCP Server Response]:", mcp_response.text)

else:

print("[GPT Response]:", response["choices"][0]["message"]["content"])

---

📜 4. Run the Demo

1. Run your MCP server:

python mcp_rpa_server.py

2. Start ngrok:

ngrok http 3333

3. Run the OpenAI client:

python openai_client.py

✅ You should see:

• GPT suggesting to call the tool.

• The MCP server triggering a Tebelorg RPA script.

• A form filled out automatically by RPA!

---

🛡 Important Notes

• You need rpa package: pip install rpa

• MCP server expects POSTs to endpoints like /automation://fill-web-form

• OpenAI API keys must be valid, and you may need GPT-4 API access (or GPT-4o).

• Tebelorg RPA uses a visible browser (Chromium/Selenium) — you need a desktop environment.

---

🚀 Bonus Tip

If you want Claude 3 instead of OpenAI, it's almost identical — you just have to send the tool definition differently using Anthropic's SDK (I can show you if you want).

---

🔥 Done!

This full-stack lets OpenAI control your real-world RPA robots via Model Context Protocol!

---

➡️

Would you like me to also show:

• A multi-tool MCP server (more than one RPA script exposed)?

• A GUI dashboard to monitor RPA tasks as they happen?

• How to secure the MCP server for production?

Tell me! 🚀