What is LINE MCP? Two integration directions compared, with a tutorial

Search for “LINE MCP” and you’ll find two things that look alike but go in opposite directions: one lets your AI agent send LINE messages, the other lets the customer service AI on LINE call your tools. Pick the wrong direction and you’ll find what you built can’t reply to customers at all. This article first explains the two directions clearly, then shows how to set up each one.

First, understand MCP: a common standard for AI to use tools

MCP (Model Context Protocol) is an open protocol that lets AI applications connect to external tools and data sources in a standard way. Once you turn “look up orders” or “check stock” into an MCP server, any AI application that supports MCP can use it, with no need to rewrite the integration for each platform.

An MCP connection has two ends: the AI that uses tools (client) and the server that provides them. The difference between the two directions of “LINE MCP” is which end LINE sits on.

The two directions of LINE MCP

Direction 1: use LINE’s official MCP server to let an agent send messages

LINE provides line-bot-mcp-server on GitHub, which wraps the Messaging API as MCP tools. Once it’s installed, you can tell your agent “send this summary to my LINE” and the agent calls the tool to send it.

1. Create a Messaging API channel in the LINE Developers Console and issue a Channel Access Token (if you don’t have an Official Account yet, see the LINE Official Account verification guide)
2. Add this server to the MCP configuration file of an agent that supports MCP (for example Claude Desktop or Cursor)
3. Restart the agent and check that tools such as push messages appear in the tool list

{
  "mcpServers": {
    "line-bot": {
      "command": "npx",
      "args": ["@line/line-bot-mcp-server"],
      "env": {
        "CHANNEL_ACCESS_TOKEN": "Your Channel Access Token",
        "DESTINATION_USER_ID": "userId of the default recipient"
      }
    }
  }
}

Note that it’s a “sending side” tool: the server runs on your computer over stdio and doesn’t receive webhooks, so the agent doesn’t know what customers reply. If your goal is for customers to ask questions on LINE and have AI answer, you need direction 2.

Direction 2: let the customer service AI on LINE call MCP tools

This direction involves more work: receiving LINE webhooks, managing each customer’s conversation state, letting AI decide when to use which tool, controlling which tools can be used in front of real customers, and handing off to a person when it can’t answer. Building it yourself from scratch is a full project; with Chatbotta’s external tools (MCP), the platform handles all of this and you only prepare the tools themselves.

Step 1: Prepare an MCP server reachable by URL

Turn the capability you want to expose (for example order lookup) into an MCP server and deploy it at a publicly reachable URL. Chatbotta connects by URL, using Streamable HTTP first and falling back to SSE if the server doesn’t support it. stdio isn’t supported; a server that only runs locally needs to be exposed over HTTP first.

No MCP server yet? If your data is small, you can put it in Google Sheets first and read it with the “Write to Google Sheets” step in Flows, then move to MCP later if needed.

Step 2: Connect your LINE Official Account

Create a project in Chatbotta and paste your Channel Access Token. The platform automatically points the channel’s webhook at itself, and messages start flowing into a unified conversation layer. For detailed steps, see Connect a LINE Official Account.

Step 3: Add an external tool (MCP connector)

In the project’s “Settings” → “Copilot and tools”, select “Add external tool” and fill in these fields:

If you choose OAuth, the platform registers a client with the server automatically according to the MCP spec and opens the browser for authorization. If the server doesn’t support automatic registration, you can enter the client ID, client secret, authorization URL and token URL manually. Tokens and secrets are stored encrypted.

Step 4: Authorize tools for the customer service AI

Once connected, tools are available only to Copilot by default, so you can look up data and build flows while designing. To let the bot that talks to real customers use them, turn on “Let the bot use it when replying to customers” and check each allowed tool in the “Bot” column of the tool list. Unchecked tools never appear in customer conversations. We recommend allowing only lookup tools at first, and turning on tools that write data after you’ve confirmed how they behave.

Step 5: When you need to identify the customer, have each customer authorize

Actions like “check my order” or “reschedule for me” require knowing who the customer is. With “Each customer authorizes”, the first time a customer uses a related tool, the AI sends an authorization link in the conversation, the customer authorizes on the source system’s own page, and from then on the AI can act on their behalf. Authorization data is stored encrypted and tied to the source system. If the customer hasn’t given all the fields a tool needs, the AI asks for them in the conversation first instead of guessing.

Step 6: Test and publish

1. Open “Test” on the right and ask questions like “Where’s my order?” to confirm the AI calls the right tool
2. Deliberately ask something the tools can’t answer and confirm it hands off to a person rather than making things up
3. After publishing, try it on your phone. The same external tools work on Messenger, Instagram DMs and website chat too, with no extra setup

Which direction should you choose?

• You want your own agent to send notifications or summaries to LINE: use LINE’s official line-bot-mcp-server
• You want customers to ask questions on LINE and have AI check your systems before answering: use Chatbotta’s external tools (MCP)
• You want both: you can use them together. The former only sends messages and doesn’t take over the webhook, so they don’t conflict
• Most questions are about fixed information like hours, prices and policies: you don’t need to write tools; an AI knowledge base is enough

For the full specs and authorization model of MCP external tools, see LINE MCP integration. If you’re still deciding whether to build your own LINE bot, start with the LINE chatbot platform comparison.

FAQ