ChatGPT + Salesforce Integration using MCP Server (Step-by-Step Guide)

👉 This is not theory — it’s fully working.

I also recorded:

🎥 Demo video

📸 Screenshots (MCP server + ngrok + Salesforce setup)

⚠️ Important Clarification (Must Read)

Let’s be honest 👇

MCP is NOT required to create a Lead in Salesforce.

You can already do it using:

• Web-to-Lead
• REST APIs
• Backend integrations
• Tools like Zapier

👉 Then why MCP?

Because when AI becomes the caller , things change.

AI needs to understand:

• What actions exist
• What inputs are needed
• How to call them properly

💡 Simple truth:

Normal apps are told what to do AI apps must be told what they can do

🧰 Tech Stack Used

🤖 ChatGPT Side

• Custom GPT → Salesforce Lead Creator

⚙️ Backend Layer

• Python FastAPI MCP-style server (Local server my laptop)
• FastMCP 3.1.1
• Endpoint: http://0.0.0.0:8001/mcp

🌐 Exposure Layer

• Created public URL using ngrok so that MCP server can access

• ngrok → public URL → local server

☁️ Salesforce Side

• Developer Edition Org
• External Connected App
• Named Credentials (External Creds, External Auth Identity Providers)
• Permission Set (I created with the name MCP access and provide External Access)
• Lead API:

{INSTANCE_URL}/services/data/v60.0/sobjects/Lead

🧠 Core Concept (Simple Breakdown)

Before setup, understand this 👇

🔹 API

Salesforce already knows how to create a Lead.

🔹 Connector

Tools like Zapier connect systems.

🔹 Backend Server

Handles logic ( /create-lead etc.)

🔹 MCP Server

👉 Used specially in the case when the Client is AI eg. in our scenarios client is Chatgpt AI, it helps AI understand:

• what actions exist
• how to call them
• what inputs to send

💡 That’s the real difference.

🍽️ Real-World Analogy

Think like this:

• 🍳 Salesforce API = Kitchen
• 🧑‍🍳 Backend = Waiter
• 🧑 ChatGPT = Customer
• 📋 MCP = Menu + Instructions

👉 Kitchen already knows cooking 👉 MCP helps place the order correctly

🛠️ Step-by-Step Setup

🔹 Step 1: Salesforce Configuration

Setup inside Salesforce:

• Connected App
• Named Credentials
• Permission Set
• API access

📸 Screenshots:

Connected App

Named Credentials

Open images in a new tab by right clicking in case not clear.

Permission Set Give access to External Creds and then add Principal User

👉 Purpose: Secure authentication + API access

🔹 Step 2: Use Lead API

Endpoint used:

/services/data/v60.0/sobjects/Lead

📌 Why Lead?

• Simple
• Easy to verify
• Real-world use case

📌 Optional fields:

• FirstName
• LastName
• Company
• Email

🔹 Step 3: Build FastAPI MCP Server

Your backend does:

• Receives request
• Validates data
• Calls Salesforce API
• Returns response

👉 This is your integration brain

🔹 Step 4: Expose via ngrok

Since backend is local:

Use ngrok to create public URL

⚠️ Note: Free ngrok URL changes on restart

🔹 Step 5: Create Custom GPT App

Create GPT: 👉 Salesforce Lead Creator

Configure:

• Enable developer mode
• Add callback URL

👉 This makes AI usable via natural language

🔹 Step 6: Add mcp_server.py

This is written in python and it handles:

• Sending requests
• Defining function schema
• Mapping inputs
• Triggering lead creation

👉 Helps understand AI → backend communication

import requests from fastmcp import FastMCP mcp = FastMCP("Salesforce Lead App") @mcp.tool( description="Create a Salesforce lead. Required fields: last_name and company. Optional fields: first_name, email, phone, rating." ) def create_salesforce_lead( last_name: str, company: str, first_name: str | None = None, email: str | None = None, phone: str | None = None, rating: str | None = None, ): payload = { "firstName": first_name, "lastName": last_name, "company": company, "email": email, "phone": phone, "rating": rating, } response = requests.post( "http://127.0.0.1:8000/create-lead", json=payload, timeout=30, ) response.raise_for_status() return response.json() if __name__ == "__main__": mcp.run(transport="http", host="0.0.0.0", port=8001)

🔹 Step 7: End-to-End Testing

Final flow:

• Select Custom GPT
• Ask: “Create a Salesforce lead for Sachin Tendulkar at TechFi Labs”
• ChatGPT processes
• Open Salesforce
• Lead is created ✅

See the above provided demo video again for end-to-end testing of ChatGPT Salesforce Integration

📸 What We are seeing now

• MCP Server running
• ngrok tunnel active
• Salesforce configs
• GPT setup

🎯 Key Learnings

1️⃣ AI + Systems = Real Power

AI alone is limited AI + Salesforce = action

2️⃣ You Already Have an Advantage

If you know:

• Objects
• APIs
• Permissions
• Flows

👉 You are already ahead

3️⃣ Start Small

You don’t need big projects

👉 One use case = huge learning

4️⃣ MCP Truth

• Not mandatory
• But powerful for AI systems

👉 Especially when AI is the caller

👨‍💻 Who Should Learn This?

🟢 Admin

Understand business flow + data

🔵 Developer

Focus on APIs + backend

🟡 Consultant

Translate business use cases

🔴 Architect

Think system-level design

⚠️ Common Challenges

You may face during ChatGPT Salesforce Integration :

• Auth issues
• Wrong callback URL
• ngrok URL changes
• Permission errors
• Payload mismatch
• Server not reachable

👉 Totally normal — this is learning

🚀 What You Can Build Next

• Create Contacts / Cases
• Query Salesforce via AI
• Build AI assistants
• Automate workflows
• AI-powered CRM tools

🔥 Final Thoughts

This was a small project…

But it showed something powerful:

AI + Salesforce is not theory. It’s buildable.

👉 You don’t need to start big 👉 Just start building

❓ FAQ

Is MCP required?

Yes — when the client is an AI-based systems , like chatgpt in our scenario

Did you use standard API?

Yes — Salesforce Lead API

Was it real?

Yes — tested on Developer Org

Biggest learning?

👉 AI becomes powerful when connected to real systems like we saw here in ChatGPT Salesforce Integration

🎯 Final Advice

If you’re:

• fresher
• admin
• developer
• career switcher

👉 Build something like this ChatGPT Salesforce Integration once

It will teach you:

• APIs
• integrations
• architecture
• real-world thinking

Checkout my other blogs articles : https://www.thetechnologyfiction.com/blog/salesforce-spring-25-new-flow-features

Recommended article to read after this : https://help.salesforce.com/s/articleView?id=platform.hosted_mcp_servers_setup.htm&type=5