MCP-Server, AI Agent, and External Tool Integration

Overview

This assignment builds an integrated AI Agent ecosystem using the Model Context Protocol (MCP).
The agent is capable of managing real-world schedules via Google Calendar and communicating
through Telegram, demonstrating practical Natural Language Understanding (NLU).

Stack used:

• n8n (local, via Docker)
• ngrok (tunnel to expose local server)
• Groq API (free LLM — llama-3.3-70b-versatile)
• Telegram Bot API
• Google Calendar API

Repository Structure

MCP-Server-AI-Agent-and-External-Tool-Integration/
│
├── README.md
│
├── screenshots/
│   │   ├── docker_running.png
│   │   ├── MCP_server_trigger_with_AI_clint_result.png
│   │   ├── ai_agent_workflow_for_telegram_and_calender.png
│   │   ├── mcp_server_workflow.png
│   │   ├── ai_clint_workflow.png
│   │   ├── other images related to results
│
└── workflows/
    ├── AI workflow.json
    ├── Server Trigger.json
    └── Telegram & Google Calendar Integration.json

1: MCP Infrastructure & Server Setup

1.1 Server Deployment

• Deployed n8n locally using Docker
• Used ngrok to expose the local server to the internet via a public HTTPS URL
• The Production URL follows the format: https://[ngrok-subdomain].ngrok-free.app/mcp/a7-server

Text Formatter Tool:

1.2 MCP Server Workflow

Created an n8n workflow acting as an MCP Server with an MCP Server Trigger node
and three internal tools:

1.3 AI Agent Client

Created a separate AI Agent workflow with:

• Groq as the LLM provider (free model: llama-3.3-70b-versatile)
• Window Buffer Memory for maintaining conversation context
• MCP Client (Tool) connected to the MCP Server Production URL

Verification — Agent using MCP tools via chat:

** AI Agent Client communication with MCP server:**

Calculator Tool:

Date & Time Tool:

Text Formatter Tool:

2: Telegram & Google Calendar Integration

2.1 Telegram Bot

• Created a Telegram Bot using @BotFather
• Added a Telegram Trigger node to the AI Agent workflow
• The agent receives messages from Telegram and replies directly in the chat

2.2 Google Calendar Tool

• Integrated Google Calendar using OAuth2 credentials in n8n
• The agent can create, read, and manage calendar events via natural language

2.3 Automated Project Scheduling

Sent the following command to the Telegram bot:

"Create a project schedule with 4 phases: Literature Review, Project Proposal,
Update Progress, and Final Presentation"

The agent automatically created 4 events in Google Calendar:

Telegram conversation:

Google Calendar result:

2.4 Interaction Verification

Asked the bot via Telegram to verify the project phases on the calendar.
The agent successfully queried Google Calendar and confirmed all events on 1st April.

Workflow Files

The exported n8n workflow JSON files are available in the /workflows folder.

Note: All credentials (API keys, tokens, OAuth) have been removed from
the exported files for security. To run these workflows, add your own
credentials inside n8n.

• [MCP Server Workflow](workflows/Server Trigger.json)
• [AI Agent Workflow](workflows/AI workflow.json)
• [Telegram & Google Calendar Integration Workflow](workflows/Telegram & Google Calendar Integration.json)

Setup Instructions

Prerequisites

• Docker Desktop
• ngrok account (free)
• Groq API key (free)
• Telegram account + Bot Token via @BotFather
• Google account with Calendar access

Running n8n locally

docker run -it --rm \
  --name n8n \
  -p 5678:5678 \
  -v ~/.n8n:/home/node/.n8n \
  -e N8N_TRUST_PROXY=true \
  n8nio/n8n

Exposing with ngrok

ngrok http 5678

Copy the HTTPS URL and update it in the MCP Client node inside n8n.

Credentials needed in n8n

References

Official Documentation

• n8n Documentation — Official n8n docs for nodes, workflows, and credentials
• n8n MCP Server Trigger Node — MCP Server Trigger setup guide
• n8n AI Agent Node — AI Agent node documentation
• n8n MCP Client Tool — MCP Client tool documentation
• n8n Telegram Trigger — Telegram Trigger node documentation
• n8n Google Calendar Node — Google Calendar integration guide
• n8n Memory Buffer Window — Window Buffer Memory documentation

Model Context Protocol (MCP)

• MCP Official Documentation — Introduction to Model Context Protocol
• MCP Specification — Full MCP specification

Tools & Services

• Docker Desktop — Container platform used to run n8n locally
• ngrok Documentation — Tunneling tool to expose local server to internet
• Groq Console — Free LLM API provider (used: qwen/qwen3-32b)
• Telegram BotFather — Official tool for creating Telegram bots
• Telegram Bot API — Telegram Bot API documentation
• Google Calendar API — Google Calendar API overview

Course Materials

• AT82.05 Course GitHub Repository — Environment setup and footprints provided by instructor
• Chaklam Silpasuwanchai and Todsavad Tangtortan, AT82.05 Artificial Intelligence: Natural Language Understanding — Assignment 7: MCP-Server, AI Agent, and External Tool Integration, Asian Institute of Technology, March 2026