LINE OA MCP Ultimate

Operate your LINE Official Account from any AI agent — through natural language.

Quick Start · What you can do · Configuration · Security · Documentation

🇹🇭 คู่มือภาษาไทย: เริ่มต้น 5 นาที · Messaging — 35 tools · LINE Shopping — 14 tools

What is this?

line-oa-mcp-ultimate is a Model Context Protocol server that lets AI agents drive your LINE Official Account through natural conversation — no code, no API knowledge required.

Manage one OA or many — a single account needs just one env var, while agencies can register every client OA in one config file and switch between them with a single command (see Configuration).

It works with Claude Cowork, Claude Code, Cursor, Codex, ChatGPT desktop, OpenClaw, Hermes, or any MCP-compatible host.

Disclaimer: This is an independent open-source project. It is not affiliated with or endorsed by LY Corporation / LINE Corp. "LINE" is a registered trademark of LY Corporation, used here in a descriptive sense.

Why use it?

LINE OA Manager works well — but it's a web UI you have to log into, click through, and operate manually. With this MCP server, the same operations become natural-language requests in your AI host of choice:

The MCP server takes care of LINE Messaging API calls, pre-flight validation, quota safety, quiet-hour warnings, and Thai-localized templates. You stay in your AI host and operate your OA the way you think about it — by intent.

🆚 vs the official @line/line-bot-mcp-server

LY Corporation publishes an official LINE Bot MCP server (Apache-2.0, marked preview). It's a clean reference implementation, and ideas from it helped shape this project — if all you need is pushing a text or Flex message to one user from an agent, it may be enough. This project targets running a real OA (or an agency's worth of them) day to day:

Based on the official repo's README and package manifest as of 2026-07 — see their repo for the latest.

Who is it for?

• 🪐 AI-first marketers who run a LINE OA from their AI workspace.
• 🏪 SMB owners who already use LINE OA as their primary customer channel and want to automate routine work.
• 🏢 Marketing agencies managing many client OAs from one dashboard.
• 💻 Developers building LINE-integrated SaaS, automations, or AI bots.

Free-tier ready — no Premium upgrade required for the core feature set.

What you can do

35 messaging tools + 14 LINE Shopping tools (v2.0, opt-in) + 4 resources + 8 guided prompts, grouped by what you actually want to do:

📨 Send messages (2 tools)

One universal send_message covers every LINE transport (reply / push / multicast / narrowcast / broadcast). Three modes: send_now, draft (for scheduling via LINE OA Manager UI), and dry_run (validate + estimate cost without sending). Message shapes: text, Flex (template or raw JSON), sticker, image, video, native LINE coupon ({ coupon_id }), and a message_json passthrough for pre-built Rich/Card messages. Plus Thai-friendly sticker search.

🎨 Rich Menus (9 tools)

Build a rich menu in one call (create + upload image + set as default), list, delete, and diagnose "why doesn't this user see my menu?". Full lifecycle too: link/unlink a menu to specific users or in bulk (auto-chunked at 500), set/clear the account default, manage rich-menu aliases (tab-switching menus), and swap the image on an existing menu. New in v2.1 — Rich Menu Studio: line_design_rich_menu_image generates the menu image itself (Thai-ready layout templates, pure-JS satori + resvg — no headless browser), lets you preview before deploying, then hands off to line_build_rich_menu.

💎 Message Design (3 tools)

Design Flex Messages from 8 Thai-localized templates or raw JSON, build Rich Messages (imagemap — tappable image regions + optional video), and Card Messages (template — buttons / confirm / carousel / image_carousel). Each returns ready-to-send JSON you hand to send_message.

🎯 Audiences (5 tools)

Build retargeting audiences from a CSV or from prior broadcast engagement, list, delete — and update an existing audience (add users, rename).

📊 Insights (5 tools)

Quick OA status snapshot (incl. chat mode), weekly markdown reports, per-broadcast engagement stats, pre-flight send-cost estimation, and narrowcast delivery progress ("did my blast finish?").

🎟 Coupons (2 tools)

Create, list, get, and discontinue native LINE coupons with line_manage_coupon, then broadcast a created coupon straight into chats as a native coupon message via send_message ({ coupon_id }). Pull redemption analytics by combining coupon details with click-audience tracking.

🔌 Webhook (2 tools)

Test your OA's webhook URL and surface LINE's signature verification result, plus set/get the webhook endpoint — quick to wire up or diagnose "why isn't my bot responding?".

👤 Operations (5 tools)

User profile lookup, follower listing, multi-OA listing and switching, and run_on_many_oas for agencies that need to run a read-only tool across all client OAs in parallel.

💻 LIFF & Token (2 tools)

Manage LIFF app lifecycle (create / update / delete / list) and verify your channel access token's validity, expiry, and scope.

🛍️ LINE Shopping (14 tools — v2.0, opt-in)

Turn your agent into a storefront operator. These register only when a MyShop API key is configured (see Configuration) — messaging-only users never see them. Powered by the MyShop Open API (X-API-KEY, no webhook/server required).

• Products (7): list, create, update, delete, change price, show/hide, and manage variants — e.g. "เพิ่มสินค้าเสื้อยืด 299 บาท สต็อก 50", "ลดราคาคอลเลกชันนี้ 15%".
• Inventory (1): set / increase / decrease stock — "เติมสต็อก V9 อีก 50".
• Orders (4): list/search (with polling for new orders via a scheduled task), get detail, fulfill (mark shipped + tracking, mark COD paid, print parcel label), and cancel — "ออกใบปะหน้าให้ออเดอร์ที่จ่ายแล้ววันนี้".
• Settlement + Checkout (2): see real income after fees, and create a checkout link to close the sale in chat — "ลูกค้าอยากได้ 2 ชิ้น สร้างลิงก์จ่ายเงิน".

📘 Full Thai guide: docs/myshop-guide-th.md — get the API key, every tool explained, real workflows, safety rules, and troubleshooting.

Resources — auto-refreshing OA snapshot, Flex template catalog, sticker catalog with mood-keyword index, and a Thai festival calendar with marketing promo patterns.

Prompts — guided workflows for daily reports, campaign building, broadcast scheduling, rich menu diagnosis, coupon campaigns, LINE Notify migration, and webhook setup.

Quick Start

Requirements

• Node.js 20 or later
• A LINE Official Account with Messaging API enabled
• A Channel Access Token

No database, no Playwright, no CDN, no public URL, no scheduler infrastructure required.

Step 1 — Get your Channel Access Token

1. Open the LINE Developers Console.
2. Pick your provider → open (or create) a Messaging API channel.
3. Open the Messaging API tab → scroll to Channel access token → click Issue.
4. Copy the long-lived token. Treat it like a password.

If you don't yet have a LINE Official Account, follow LINE's getting started guide.

Step 2 — Add the MCP server to your AI host

Paste the following block into your host's MCP config file (see table below):

{
  "mcpServers": {
    "line": {
      "command": "npx",
      "args": ["-y", "line-oa-mcp-ultimate"],
      "env": {
        "LINE_CHANNEL_ACCESS_TOKEN": "YOUR_TOKEN_HERE"
      }
    }
  }
}

Where to find the config file:

⚠️ Don't replace existing MCP servers. Add the "line" entry alongside any existing entries inside mcpServers.

Step 3 — Restart your AI host

Quit completely (Cmd+Q on macOS) and reopen. On first start the host runs npx -y line-oa-mcp-ultimate, downloads the package from npm, and caches it for subsequent starts.

Step 4 — Verify

In a new chat, try:

Show my LINE OA status

or in Thai:

ดูสถานะ OA หน่อย

You should see a health card with friend count, monthly quota, webhook status, and the default rich menu. That's it — you're now operating your LINE OA from your AI agent.

🛍️ Want LINE Shopping too? (optional)

Add LINE_MYSHOP_API_KEY to the same env block (get it from oaplus.line.biz → Settings → API keys, Admin role). The full config looks like this:

{
  "mcpServers": {
    "line": {
      "command": "npx",
      "args": ["-y", "line-oa-mcp-ultimate"],
      "env": {
        "LINE_CHANNEL_ACCESS_TOKEN": "YOUR_TOKEN_HERE",
        "LINE_MYSHOP_API_KEY": "YOUR_MYSHOP_API_KEY_HERE"
      }
    }
  }
}

Restart your host — that unlocks 14 shopping tools (products, inventory, orders, parcel labels, settlements, checkout links). Without the key, the server stays messaging-only. See LINE Shopping, the full Thai MyShop guide, and Image hosting.

Configuration

Single-OA mode (default — for most users)

That's the Step 2 snippet above. One environment variable, one OA. Suitable for individual creators, course makers, and most SMB owners.

Multi-OA mode (for agencies)

If you manage several LINE OAs (your own plus client accounts), create a JSON config file at ~/.line-mcp/config.json — the server auto-discovers it, so there's nothing to clone or build:

mkdir -p ~/.line-mcp
nano ~/.line-mcp/config.json     # or: code ~/.line-mcp/config.json

📘 Full walkthrough (two placement options + security notes) in docs/multi-oa-setup-th.md. A ready-to-edit template lives in config.example.json.

Fill in one entry per OA:

{
  "default_oa": "main",
  "oas": {
    "main": {
      "channel_access_token": "TOKEN_1",
      "display_name": "Main OA",
      "region": "TH",
      "myshop_api_key": "MYSHOP_KEY_1"
    },
    "client_a": {
      "channel_access_token": "TOKEN_2",
      "display_name": "Client A — Bangkok Cafe"
    }
  }
}

🛍️ LINE Shopping (v2.0): add "myshop_api_key" to any OA to enable the 14 shopping tools for it (get it from oaplus.line.biz → Settings → API keys, Admin role). In single-OA mode, set the env var LINE_MYSHOP_API_KEY instead. OAs without a key simply keep the messaging tools — the shopping tools never appear and never error.

Then drop the LINE_CHANNEL_ACCESS_TOKEN env var from your MCP config — the server discovers ~/.line-mcp/config.json automatically.

Resolution order:

1. LINE_MCP_CONFIG env var (custom path)
2. ~/.line-mcp/config.json
3. LINE_CHANNEL_ACCESS_TOKEN env var (single-OA fallback)

Switching the active OA at runtime

Ask your AI host:

"Switch to client_a"
→ line_use_oa({ oa_id: "client_a" })

Or override per-call with the oa parameter on any tool.

Advanced — HTTP transport (self-hosted)

For agencies running a shared remote instance, the server also supports Streamable HTTP. See docs/http-transport.md.

🖼️ Image hosting

Every tool that uses an image — broadcast image, rich menu, card, imagemap, flex, and LINE Shopping products (v2.0) — needs a public, direct HTTPS URL to a JPEG/PNG. LINE fetches the image from that URL, so it must be directly accessible (not a preview page). MyShop has no binary upload endpoint, so host the image first and pass its URL.

• ✅ Recommended: Cloudflare R2, AWS S3 (public bucket), raw.githubusercontent.com, Cloudinary, ImageKit — stable, direct URLs.
• ⚠️ Avoid: Google Drive / Dropbox share links — they return an HTML preview page, not a direct image, so LINE can't read them reliably.

Documentation

🔐 Security

This server operates with live channel access tokens — whoever holds the token is your shop. Treat your MCP config like a password store.

• Report vulnerabilities privately via GitHub Security Advisories — see SECURITY.md for scope, response times, and what to include. Please don't open public issues for security problems.
• Thai threat-model guide for shop owners: docs/security-th.md — what a leaked token can do, per-OA token isolation, and config-file permissions.
• HTTP transport requires auth beyond localhost: binding to any non-loopback host (e.g. 0.0.0.0) requires MCP_HTTP_TOKEN — every request must then carry Authorization: Bearer <token>. Without the token the server refuses to expose itself beyond 127.0.0.1. Details in docs/http-transport.md.

Versioning

This project follows Semantic Versioning. Current release: v2.1.0 "Trust Release" — full history in CHANGELOG.md.

Latest changes (v2.1.0): Rich Menu Studio (line_design_rich_menu_image — generate rich-menu images with no headless browser, preview before deploy), HTTP bearer auth (MCP_HTTP_TOKEN), a published security policy (SECURITY.md), CI/CD with npm provenance, and a much larger test suite with a coverage gate enforced in CI. See CHANGELOG.md.

Previously (v2.0.0): +14 LINE Shopping tools via the MyShop Open API — products, inventory, orders (incl. parcel labels), settlements, and checkout links. API-key only, no webhook/server, fully opt-in (registers only when a MyShop key is set) and non-breaking on the 34 messaging tools. See RELEASE_NOTES_v2.0.0.md.

Previously (v1.1.0): +10 tools completing LINE's token-only Messaging API surface (rich-menu lifecycle, Rich/Card message builders, audience update, webhook set/get, narrowcast progress, token check). See RELEASE_NOTES_v1.1.0.md.

Roadmap

• v1.x — Full token-only Messaging API coverage: outbound + rich-menu lifecycle + Rich/Card message builders + audiences + insights + coupons.
• v2.0 — ✅ LINE Shopping (MyShop Open API) shipped — products, inventory, orders, parcel labels, settlements, and checkout links (API-key only, no webhook). See RELEASE_NOTES_v2.0.0.md.
• v2.1 — ✅ "Trust Release" shipped — Rich Menu Studio (image generation + preview), HTTP bearer auth, security policy + hardening, CI/CD with provenance, coverage-gated test suite. See CHANGELOG.md.

Future direction is guided by real-world usage and community feedback.

Contributing

PRs are welcome. Please open an issue first to discuss substantial changes.

The one rule: a new tool ships with new tests. Every tool is exercised through a real in-memory MCP client with the LINE API mocked at the fetch boundary (see tests/helpers/) — no live token needed to run the suite.

npm ci
npm run typecheck       # strict TypeScript, no `any`
npm test                # vitest run
npm run test:coverage   # coverage — ≥80% lines, enforced in CI
npm run build

When contributing:

• Tool names: line_{action}_{resource} (snake_case, prefix line_).
• Tool descriptions: English (for LLMs) — Thai examples encouraged.
• User-facing strings: Thai-friendly (centralized in src/i18n/th.ts).
• Pre-flight validation on every send tool.
• Quota Guardian + Quiet Hours on every consumption tool.
• Backward compatibility is non-negotiable — never rename or remove existing tools, params, response fields, env vars, or config keys.

PR expectations: CI must be green (test workflow — typecheck, tests with coverage, build on Node 20 & 22), and behavior changes come with a matching CHANGELOG.md line under Unreleased.

Releasing (maintainers)

1. Move the Unreleased section of CHANGELOG.md to the new version and bump package.json.
2. git tag vX.Y.Z && git push origin vX.Y.Z.
3. The release workflow runs the full gate (typecheck + tests + build), publishes to npm with --provenance, and creates a GitHub Release from the matching CHANGELOG section. It needs one repo secret: NPM_TOKEN (npm granular access token with publish rights on this package).

Author

Dr. Wasin Treesinthuros (พี่โต / Dr. Wasin / #จารโต)

• CEO at Innovation Vantage
• Founder at ForgeAI — AI Transformation Studio
• LINE Certified Coach 2026
• AI Transformation educator, SaaS builder, university lecturer

This project is part of a broader mission to bring production-grade AI automation to every Thai practitioner — regardless of technical background.

GitHub: @wasintoh

License

MIT © Dr. Wasin Treesinthuros — Innovation Vantage / ForgeAI