Zendesk MCP Server (Model Context Protocol)

This project is a lightweight, AI-native MCP (Model Context Protocol) server that integrates with Zendesk's REST APIs. It allows GPT-based AI agents (e.g. OpenAI, LangChain) to fetch real-time customer and organization context dynamically.

Features

• Accepts ticket_id, user_id, or organization_id

• Fetches user, org, and ticket context from Zendesk

• Returns:

  • summary: human-readable LLM-friendly summary

  • prompt_context: single-line LLM embedding string

  • context: structured blocks (text, list)

  • prompt_guidance: usage instructions and few-shot examples

• Exposes:

  • /context: main context API

  • /meta: MCP schema metadata

  • /function-schema: OpenAI function-compatible definition

• Fully Dockerized and deployable

• Compatible with GPT-4 function calling

Related MCP server: Zendesk MCP Server

Getting Started

1. Clone and install dependencies

2. Set up .env

3. Run Locally

Visit:

• http://localhost:3000/context

• http://localhost:3000/meta

• http://localhost:3000/function-schema

Docker Support

Build Image

Run Container

Function Calling with OpenAI (Example)

See openai-client.js for an example where:

• GPT-4 automatically detects and calls get_ticket_context

• The function calls your local MCP server

• GPT writes a natural reply using the returned context

Simulating a Full Chat Conversation

What you've tested so far is GPT-4 calling your MCP server using function calling, which works. Now you want to simulate a full conversation where:

A user asks something natural like:

“Can you give me context for ticket 12345?”

• GPT-4 figures out it needs to call get_ticket_context

• GPT-4 calls your MCP server automatically

• GPT-4 uses the result to reply in a natural, chat-style response

Let’s build exactly that — your own OpenAI Agent Loop that mimics how GPT-4 with tools (functions) will behave in production.

✅ Step-by-Step: Full Chat-Based OpenAI Agent with Function Calling

✨ Final Output Looks Like:

What This Script Does:

• Sends a natural user message to GPT-4

• GPT-4 detects your function, calls it with a ticket_id

• You send that to your MCP server

• Feed the MCP server’s context result back to GPT

• GPT-4 writes a human-style response using the result

Web Chat Interface + OpenAI Router API

To demonstrate end-to-end usage with real input/output, this project includes:

1. /chat API endpoint (openai-router.js)

A Node.js API that accepts natural language messages, detects intent using GPT-4 + function calling, and uses the MCP server to fetch data and compose replies.

🔧 .env additions:

▶️ Run the API:

This starts a server at http://localhost:4000/chat

2. chat-ui.html

A simple HTML frontend to type user prompts and see AI-generated responses with Zendesk context.

🧪 Example Questions:

• Who is the user for ticket 12345?

• Tell me about organization 78901

• How many tickets has user 112233 opened?

💬 Usage

• Open chat-ui.html in a browser

• Ensure the /chat endpoint is running with CORS enabled

• Ask questions and see the result appear naturally

🔐 Note

Make sure you install and enable CORS in openai-router.js:

Future Enhancements

• LangChain tool compatibility

• Redis caching layer

• Rate limiting

• More context types: /orders, /billing, /subscriptions

License

MIT

Author

Your Name — @yourhandle