LINE OA MCP Ultimate

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

🛠️ Fixed in 2.2.1: run the MCP on your own server and point any AI agent at one URL — it now holds up through real sessions and full broadcasts (no more instant "Internal server error" in HTTP mode), and hosted pictures now cache properly, so thousands of customers opening your broadcast at once barely touch your machine. Thanks to a production field report by Norapat Limpagan! 🙏

🖼️ New in 2.2: attach a picture, say "send this as a Rich Message" — and it appears in your customers' chats. No hosting. No image resizing. No setup.

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

🇹🇭 คู่มือภาษาไทย: เริ่มต้น 5 นาที · Messaging — 37 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:

| Instead of clicking through LINE OA Manager... | Just say... | |---|---| | Open broadcast composer → pick audience → write text → preview → schedule | "Send a Mother's Day promo to my Bangkok-based 25-44 female audience" | | Resize the artwork to 5 sizes → find somewhere to host it → upload → paste URLs into a Rich Message | "ส่งรูปนี้เป็น Rich Message หาลูกค้าทุกคน" (attach the picture — that's the whole job) | | Navigate Insight → Delivery → pick date range → screenshot | "Summarize last week's performance" | | Open Audience → create from upload → upload CSV → wait | "Create an audience from this CSV: customers_last_month.csv" | | Open Rich Menu → design tabs → upload images → set as default | "Build a Mother's Day rich menu and set it as default" | | Open Coupon → check usage tab → cross-reference clicks | "How many people redeemed my Mother's Day coupon?" |

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:

| | line-oa-mcp-ultimate (this project) | official @line/line-bot-mcp-server | |---|---|---| | Tools | 51 (37 messaging + 14 shopping) + 4 resources + 8 guided prompts | 12 (push/broadcast text & Flex, profile, quota read, rich-menu basics, follower IDs) | | LINE Shopping (MyShop) | ✅ 14 opt-in tools — products, stock, orders, parcel labels, settlements, checkout links | ❌ | | Rich Message images | ✅ zero-setup (v2.2) — attach a picture, everything else is automatic and verified delivered-able before "success" | ❌ bring your own hosted URLs | | Multi-OA | ✅ one config file for all client OAs + runtime switch (line_use_oa) + fan-out (line_run_on_many_oas) | ❌ one OA per server instance | | Quota / quiet-hours guardrails | ✅ pre-flight Quota Guardian, 22:00–08:00 quiet-hours warnings, dry_run cost estimates | ➖ quota read tool (get_message_quota); no pre-flight guardrails | | Thai i18n | ✅ every user-facing error/warning in Thai, plus full Thai guides | ❌ English (Japanese README available) | | Rich-menu image generation | ✅ pure-JS satori + resvg (~10 MB, no browser), with a preview-before-deploy loop | ✅ Marp + Puppeteer (downloads headless Chromium), generate-and-upload in one shot | | Transports | stdio + Streamable HTTP with bearer-token auth (MCP_HTTP_TOKEN) | stdio (npx / Docker) | | Runtime footprint | browser-free — no headless Chromium download | depends on puppeteer + @marp-team/marp-core for rich-menu images |

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

37 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 (5 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. New in v2.2: Rich Messages need zero hosting — attach any image and it's resized, hosted, and verified for you (line_prepare_image + line_image_host_status).

🎯 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:

| Host | Where to add it | |---|---| | Claude Cowork | Settings → Developer → Local MCP servers → Edit Config | | Claude Desktop (macOS) | ~/Library/Application Support/Claude/claude_desktop_config.json | | Claude Desktop (Windows) | %APPDATA%\Claude\claude_desktop_config.json | | Claude Code | Run claude mcp add line or edit ~/.config/claude/mcp.json | | Cursor / Codex / ChatGPT desktop | See your host's MCP setup docs — same JSON shape |

⚠️ 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 and always-on servers, the server also supports Streamable HTTP — run one instance per agent/teammate (the active OA switched by line_use_oa is shared per instance). See docs/http-transport.md.

🖼️ Images & Rich Messages — zero setup (v2.2)

Before 2.2, sending one Rich Message meant: resize your artwork into five exact sizes, find hosting, upload, paste URLs — the step where most shop owners simply gave up. Now the whole workflow is:

Attach a picture. Say "send this as a Rich Message". Done.

What you get, without doing anything:

• 📐 Every size LINE requires — prepared automatically.
• 🌍 Your image made reachable for LINE — automatically, wherever you run: your laptop, a server, anywhere. Nothing to sign up for, nothing to configure.
• ✅ Verified before "success" — the tool checks that LINE can actually pull every size before it ever tells you the send worked. The classic "API said success but customers see a blank image" problem is gone.
• 🧰 Never a dead end — on locked-down networks where automation isn't possible, you instantly get a ready-made package + 1-minute Thai instructions instead of an error.

One honest rule to remember: LINE collects the image when each customer first opens the message — so keep your machine on until your audience has seen it (the built-in 24-hour window covers a normal broadcast day). Once opened, the image lives on LINE's side permanently.

Full guide + troubleshooting (Thai): docs/image-hosting-th.md

Prefer hosting yourself (agencies, long-running campaigns)? Everything still works the classic way — every image tool accepts a public, direct HTTPS URL to a JPEG/PNG:

• ✅ 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. (The source_url input rewrites these to direct-download form automatically.)

Documentation

| Doc | What's in it | |---|---| | docs/quickstart-th.md | Thai-language quick start (5-minute walkthrough) | | docs/messaging-guide-th.md | 📨 Messaging full guide — Thai. The messaging tools explained, transports & send modes, real workflows, safety limits, and troubleshooting | | docs/myshop-guide-th.md | 🛍️ LINE Shopping (MyShop) full guide — Thai. Get the API key, all 14 tools explained, real end-to-end workflows, safety rules, and troubleshooting | | docs/clients-setup-th.md | Per-host MCP setup — Cowork / Claude Desktop / Cursor / Codex (Thai) | | docs/image-hosting.md · docs/image-hosting-th.md | 🖼️ Images & Rich Messages with zero hosting (v2.2) — EN + Thai: how it works, the first-view rule, troubleshooting | | docs/multi-oa-setup-th.md | Multi-OA configuration guide (Thai) | | docs/cowork-local-test-th.md | Test a local build in Claude Cowork before publishing (Thai) | | docs/http-transport.md | Streamable HTTP transport for self-hosted / remote use | | SECURITY.md · docs/security-th.md | Security policy (EN) · threat-model guide for shop owners (Thai) | | examples/ | Runnable examples |

🔐 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: v2.2.2 (docs-only republish of 2.2.1) — full history in CHANGELOG.md.

Latest changes (v2.2.1): HTTP transport fixed for real MCP clients — every request now gets its own server instance (overlapping requests no longer 500), non-POST methods answer a clean 405, and self-hosted images serve immutable cache headers + ETag/304 so a reverse proxy or CDN can absorb an entire broadcast wave. Reported from production by Norapat Limpagan — thank you! See CHANGELOG.md.

Previously (v2.2.0): Image Hosting Layer — attach a picture and line_prepare_image resizes it to every size LINE needs, hosts it automatically (own server / quick tunnel / manual-package fallback), and verifies every size is reachable before ever reporting success; line_image_host_status gives a read-only health check.

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.
• v2.2 — ✅ "Zero-Hosting Release" shipped — attach a picture, say send: auto-resize to all 5 Rich Message sizes, automatic hosting (own server / quick tunnel / manual package — nobody dead-ends), every size verified reachable before success. 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