LINE Workflow Automation with Claude Code & MCP

Course Details

A 2-day onsite hands-on workshop, 6 hours per day (12 hours in total). Every workshop across both days is woven around one scenario: a simulated company, "Siam Smart Trade", a Thai trading company that communicates mainly through LINE. On Day 1 learners practice Claude Code with this company's sales data, and on Day 2 they take on the role of the engineer building automation for the HR, operations and management teams. It closes with a Group Challenge where each team picks a real problem from their own organization for the AI agent to solve and then presents it. All data is simulated (a database seed file is provided) and unrelated to any real company.

Day 1: A Complete Claude Code Foundation for Beginners - From Install to Automation

Section 1: Getting to Know AI Agents and the Course Goals

• From AI chat to AI agent: the difference that changes how you work, and why Thai organizations should start their auto workflow with LINE
• Getting to know Claude Code and GPT Codex: what a CLI-based AI agent is and what it can do
• An overview of the system you will finish building on Day 2, so you can see the destination of every skill learned today
• Opening activity: survey the real workflows the organization currently runs on LINE and record them as a problem bank for the Day 2 Group Challenge

Section 2: Getting Started with Claude Code - Install, Instruct and Control

• Install Claude Code on Windows, log in to your account and check the machine is ready
• How it works: sessions, the project folder and the instruct-review-approve cycle
• Your first commands: creating, reading and editing files and running commands through AI
• Permissions and safety: approving file edits and command runs, Plan Mode (plan before acting), pressing ESC to stop mid-task, and redirecting the work
• The key idea of the day: the AI does the work, you direct and accept it

Section 3: Key Commands, Context Management and CLAUDE.md

• Frequently used commands: /help, /clear, /compact, /model, /mcp, /init and resuming work (continue and resume)
• What context is, why a long session makes the AI lose track, and how to manage it with /clear and /compact
• CLAUDE.md, the project memory: set rules such as answer in Thai, no semicolons, use the specified library
• Techniques for good instruction prompts: state the goal, scope, technical requirements and a verifiable result, with side-by-side good and bad prompt examples from real work

Section 4: Workshop 1 - Your First Project with Claude Code

• Build your first project from scratch: have AI create a web page summarizing Siam Smart Trade with a Chart.js graph
• Practice the full cycle: writing prompts, using Plan Mode to review the plan before approving, and reviewing and re-directing the work until satisfied
• Create a CLAUDE.md to set project rules, then verify the rules really affect the next task
• This skill set is the heart of building the Webhook Server and reports on Day 2

Section 5: Model Context Protocol - The Gateway from AI to Your Systems

• What MCP is: the standard for connecting AI to external systems, its architecture and the ecosystem of MCP servers you can use right away
• Adding MCP with the claude mcp add command, checking status with /mcp, and configuration scope (project only or whole machine)
• Introducing the two MCPs used throughout the course: the Postgres MCP (read side) and the LINE Bot MCP Server, with a deep dive into all 12 of its LINE tools

Section 6: Workshop 2 - Connect MCP to a Database and Have AI Analyze in Thai

• Import Siam Smart Trade's simulated sales database (seed file provided, 30 days of history)
• Connect the Postgres MCP to Claude Code yourself, then practice analysis step by step: explore the table structure, summarize yesterday's sales by branch and channel, and analyze the 30-day trend
• Finish with a task to find the abnormal sales record hidden in the data (anomaly detection)
• This database is used again in the Day 2 Capstone

Section 7: Workshop 3 - From Manual to Automatic: Your First Report Script and Task Scheduler

• Key idea: ad-hoc work uses MCP on demand, while recurring work uses a script plus a schedule, and Claude Code writes both kinds of script for you
• Have AI write a Node.js script that summarizes yesterday's sales into a report file, testing the dry-run concept before going live
• Schedule it to run automatically with Windows Task Scheduler and test an immediate run
• This same skeleton becomes the LINE report in the Day 2 Capstone

Section 8: GPT Codex Cross-check and Day 1 Wrap-up

• Get to know the Codex CLI: install it, configure MCP (the config.toml file), then run the same prompts as Workshop 2 and compare the results and approach with Claude Code
• Principles for writing tool-neutral prompts that work with any AI agent
• Summarize all of Day 1's skills against the table showing where each will be used tomorrow
• Prepare for Day 2: register a LINE Developers account, form teams of 3-4, create a test LINE group and verify the ngrok installation

Day 2: Apply Every Skill to Build a Full LINE Auto Workflow System

Section 9: The LINE Platform for Organizational Work

• An overview of the LINE Official Account and Messaging API: what it can and cannot do, and each plan's message quota
• The key limitation: the Messaging API cannot read past chat history, which sets up the afternoon's Webhook plus database architecture
• The overall architecture of today's system: Webhook (inbound), PostgreSQL (memory), Postgres MCP (read) and LINE Bot MCP (send back)

Section 10: Workshop 4 - Set Up the LINE OA and an HR Notification Bot

• Create a Provider and Messaging API Channel in the LINE Developers Console, configure the OA Manager (enable Webhook, disable auto-response, allow the bot into groups) and issue a Channel Access Token safely
• Connect the line-bot MCP to Claude Code (the claude mcp add skill from Day 1) and send your first message into your own LINE with a single Thai prompt
• Play the HR role: send individual meeting invitations, pull the follower list with get_follower_ids, broadcast a holiday announcement to every employee and check the quota with get_message_quota
• Flex Message: design an internal job posting with buttons, test it in the Flex Message Simulator, and build an employee-services Rich Menu with create_rich_menu

Section 11: Workshop 5 - Group Chat Recorder: Capture Group Chats into a Database

• Invite the bot into each team's test LINE group, then have Claude Code build a Webhook Server (Node.js, Express, @line/bot-sdk) that receives group messages and saves them to PostgreSQL
• Use the Day 1 skills in full: write a prompt with clear requirements, use Plan Mode to review the plan, and review the AI-generated code before approving
• Cover signature verification, deduplicating data when LINE retries, and always responding 200; expose the system to HTTPS with ngrok and set the Webhook URL
• Have group members chat for real on a simulated topic (planning the operations team's work) and verify the messages flow completely into the database

Section 12: Workshop 6 - AI Summarizes the Group Meeting and Tracks Tasks, Posted Back to the Group

• Use the Day 1 MCP database-analysis skills on the team's real chat data: summarize what the group discussed, who raised which points, and analyze group behavior
• Find the appointments and tasks people committed to and turn them into a task-tracking table
• Highlight: have AI summarize the meeting with the outstanding task list and post it back to the group as a daily Flex Message report everyone sees at once

Section 13: Workshop 7 (Capstone) - Executive Daily Report Delivered Automatically Every Morning

• Extend the report script from Day 1 Workshop 3 into a complete system: have Claude Code build a daily-report script that queries yesterday's sales from the database imported on Day 1
• Assemble it into an executive Flex Message report with a red alert section when an anomaly is found
• Test with a dry run, check the JSON, then send it live to the group, and schedule it to run automatically every morning at 08:00 with Task Scheduler
• By the end of the workshop every team has an automated report that runs on its own with no one issuing a command

Section 14: Group Challenge - Solve a Real Organizational Problem and Present

• Each team picks one real problem from the problem bank collected since the start of the course, such as a notification system they wish they had, a report done by hand every day, or a data question management asks often
• Have the AI agent build a prototype within the time limit
• Present it to the room for 3 minutes per team and receive guidance on real-world adoption from the instructor

Section 15: Governance, Privacy and Course Wrap-up

• Governance and privacy: obtaining consent before capturing group chats, using a read-only user with a real database, sensitive data that must not leave the organization, and token care
• Ways to go further after the course: alert scripts for anomalies only, weekly HTML reports, building your organization's own MCP server and moving the system to the cloud
• Closing Q&A, and distribution of all prompt documents with a checklist for real-world organizational adoption