For developers and technical teams

Connect customer chats to your MCP tools

Chatbotta handles conversations on LINE, Messenger, Instagram and website chat. You build your tools as an MCP server and plug it in. No webhooks, message formats or conversation state to maintain.

Get started free Read the help docs

Why it matters

Where DIY integrations get painful

Channels, data and permissions: each one is something to maintain.

Wiring up social platforms takes real work

LINE webhooks and reply tokens, plus separate message formats for Messenger and Instagram. Just receiving messages reliably is a project before you’ve built anything useful.

The model is easy. Your data isn’t.

For an LLM support agent to look up orders, check stock or read internal docs, you build every integration yourself. MCP is already the standard; what’s missing is the link to customer conversations.

Tool permissions are hard to control

Not every tool you give the AI belongs in real customer conversations. You need fine-grained control over which tools go live and which stay internal for design work.

Two directions of LINE MCP

How is this different from LINE’s official MCP server?

Search “LINE MCP” and you’ll find two complementary tools that work in opposite directions.

AI agent → LINE

LINE’s official line-bot-mcp-server

Who uses the tools: A developer’s own AI agent (Claude Desktop, Cursor, etc.)
What it does: The agent calls the Messaging API through MCP: push text or Flex messages, look up user profiles, manage rich menus
Good for: Internal automation, notifications the agent sends on its own
Note: Runs locally over stdio; doesn’t receive customer messages

LINE customer chat → your MCP tools

Chatbotta MCP external tools

Who uses the tools: Customer service AI talking to real customers
What it does: A customer asks on LINE; the AI calls your MCP tools to look up an order or change a booking, then replies in plain language
Good for: Customer support, order and booking lookups, customer self-service
Note: The platform handles webhooks, conversation state and handoff to a person. Messenger, Instagram and website chat share the same external tools

Want an agent to send LINE messages for you? Use LINE’s official line-bot-mcp-server. Want customers to ask questions on LINE while the AI calls your systems to answer? Use Chatbotta. For a full comparison, see the LINE MCP tutorial.

Getting started

Live in three steps

The conversation layer is ready-made. You focus on the tools.

Connect your channels

Paste your LINE Official Account’s Channel Access Token (Messenger, Instagram and website chat take a few similar steps). Messages start flowing into one conversation layer, and the platform handles webhooks and conversation management.

Add MCP external tools

Install a ready-made tool from “Apps” in one click, or go to “Settings” › “Copilot and tools”, select “Add external tool” and point it at your own MCP server.

Authorize tools for the customer service AI

Turn on and check each tool you want in real customer conversations. When a customer asks “Where’s my order?”, the AI calls the tools you’ve allowed to find the answer.

Features

What you get

From one-click installs to custom external tools, with permissions you control.

One-click install from “Apps”

Install the booking system or the Cloudbeds hotel system from “Apps” in one click, and read and write Google Sheets, with no config files or runtime to manage. You can add custom MCP servers manually too.

Custom MCP external tools

Add an external tool (what other AI products call an MCP connector) to plug in your own MCP server: internal systems, order databases, any tool.

Authorize each tool

New external tools are available only to Copilot by default. To use one in real customer conversations, turn that on separately and check each tool you allow.

Copilot can use them too

While you design flows, Copilot can call the external tools you’ve added, so you can build flows and integrations by chatting.

One conversation layer, every channel

LINE, Messenger, Instagram and website chat messages arrive in the same conversation layer. Connect your MCP tools once and all four channels can use them.

Customers authorize once, the AI acts for them

After a customer authorizes once, the AI can look up or change their own orders for them. Authorization data is stored encrypted and bound to the source system, so it can’t be misused across systems.

Asks for missing details

If a customer leaves out fields a tool needs, the AI asks for them in the chat and continues automatically once they’re filled in. It never guesses and submits.

Integrate without MCP

If your data lives in Google Sheets, the “Write to Google Sheets” flow step reads and writes it directly. Simple lookups don’t need an MCP server.

Technical specs

Custom external tool specs

An external tool is an MCP connector. Check these before adding your own MCP server.

Transport: Remote MCP servers, connected by URL. Streamable HTTP first, with automatic fallback to SSE if the server doesn’t support it. stdio isn’t supported; expose a local server over HTTP first.
Authentication: None (public server), API token / bearer token, OAuth (with dynamic client registration and PKCE; if the server doesn’t support automatic registration, enter the client ID, secret, authorization URL and token URL manually), or per-customer authorization (OAuth per user).
Credential storage: Tokens, client secrets and customer authorization data are encrypted with AES-256-GCM. Credentials in tool responses are masked before anything is written to logs.
Tool list: Fetched once after connecting and cached, not re-queried in every conversation. When the server’s tools change, refresh the tool list to sync.
Tool authorization: Tools are open only to Copilot by default. To let the customer service AI use them, turn that on separately and check each tool. Only tools already open to Copilot can be checked.

Fields for a new external tool

Fill them in under “Settings” › “Copilot and tools” › “Add external tool”. No config file needed.

Name Order lookup

Server URL https://mcp.example.com/mcp

Purpose Look up order status and shipping progress (the AI uses this to decide when to use the tool)

Authentication API token / bearer token

API token Paste the token (stored encrypted)

You draw the flow, the AI uses it

Your flows can be AI tools too

Not just external systems: the logic you design can be called by the AI as well.

Open any flow from the visual designer to the AI as a tool with one click. When a customer asks “Do I get a discount if I buy this way?”, the AI runs your decision flow with the customer’s numbers and answers with the result.

You set the rules in the designer. The AI understands the question, passes in the parameters and explains the result. For details, see Features: flows as tools.

Checked before going live

Tool flows are checked automatically before going live: they won’t send messages on their own, always return a value and have no infinite loops.
Parameters are used once

Parameters from the AI are used only for that one calculation and aren’t written to customer data.
You own the rules, the AI does the talking

That split keeps answers reliable. Discounts, eligibility and fees always follow the flow you drew.

Typical scenario

The AI checks your internal systems

LINE is the example; Messenger and Instagram DMs work exactly the same way.

Build order lookup as an MCP server (or start with order data in Google Sheets, read by a flow’s spreadsheet step)
Under “Settings” › “Copilot and tools”, add an external tool with the server URL and authentication
Turn on access for the customer service AI and check the lookup tools you allow
A customer asks “Where’s my order?” on LINE, or the same thing in an Instagram DM. The AI calls the same tool, gets the status and replies in plain language; anything it can’t answer can still go to a person

The same external tools work on all four channels: LINE, Messenger, Instagram and website chat. Knowledge questions go to the knowledge base, with no tools to write.

FAQ

Frequently asked questions

What is MCP (Model Context Protocol)?

MCP is an open protocol for connecting AI applications to external tools and data sources in a standard way. Build a tool as an MCP server once, and any AI application that supports MCP can use it, with no per-platform rewrite.

How is Chatbotta different from LINE’s official line-bot-mcp-server?

They point in opposite directions and complement each other. LINE’s official line-bot-mcp-server lets your own AI agent (such as Claude Desktop or Cursor) call the Messaging API through MCP to push messages to LINE users. Chatbotta lets the customer-facing AI on your LINE Official Account call your MCP tools mid-conversation to look things up and get things done, while the platform handles webhooks and conversation management.

Can I connect an MCP server I built myself?

Yes. Besides one-click installs in “Apps”, you can add an external tool under “Settings” › “Copilot and tools” to plug your own MCP server into a project for Copilot or the customer service AI.

What transports and authentication do custom MCP external tools support?

Remote MCP servers connected by URL: Streamable HTTP first, with automatic fallback to SSE if the server doesn’t support it; stdio isn’t supported. Authentication options are none, API token / bearer token, OAuth (with dynamic client registration or a manually entered OAuth app), and per-customer authorization. All credentials are stored encrypted.

Can Messenger and Instagram use MCP tools too?

Yes. LINE, Facebook Messenger, Instagram DMs and website chat all share one conversation layer, and MCP external tools attach to the project, not to a single channel. Set them up once and the customer service AI on all four channels uses the same authorized tools, with no per-channel setup.

Are MCP tools exposed to customers by default?

No. New external tools are available only to Copilot by default. For the customer-facing AI to use one, turn that on separately and check each tool you allow. Unchecked tools aren’t used in customer conversations.

Can I integrate without MCP?

Yes. If your data lives in Google Sheets, the “Write to Google Sheets” flow step reads and writes the spreadsheet with no MCP server. Use MCP external tools when the AI needs to query your own systems on its own.

Can the AI change orders for customers?

Yes. After a customer authorizes once in the chat, the AI can look up or change their own orders, such as rescheduling or updating contact details. Authorization data is stored encrypted and bound to the source system, and the AI can only touch that customer’s own data. If required fields are missing, the AI asks for them before running the tool.

Can the AI use flows from the flow designer?

Yes. Open any flow you’ve drawn in the visual designer to the AI as a tool with one click. Before it goes live, the system automatically checks that the flow won’t send messages on its own, always returns a value and has no infinite loops. Parameters from the AI are used only for that one calculation and aren’t written to customer data.

Never leave a customer waiting

Try it for 14 days with real customers. We’ll help set up your knowledge base and flows, and can extend the trial if you need more time. Rather talk first? Book a 30-minute demo.

Start for free Book a free demo

Trial: 14 days, all features
Onboarding: We help set up your knowledge base
Payment: No credit card, no automatic charges
Contract: Monthly billing, no long-term commitment