Line.OpenApi.Tools 1.3.0

English | 日本語

Line.OpenApi.Tools — CLI / MCP tool for LINE

A dotnet global tool (command name line) for operating the LINE Platform from your local machine. Built on top of the Line.OpenApi.* client libraries, it exposes the same functionality both as CLI subcommands and as MCP server tools.

• CLI — run from a terminal by a human (line message push ...)
• MCP server — called by AI agents (Claude Desktop / Claude Code, etc.) via line mcp over stdio

Both share the same service layer, so behavior is identical.

Features

Area	Capabilities
A. Token management	Issue (v2.1 JWT / stateless), verify, and revoke channel access tokens
B. Message send & bot lookup	push / multicast / broadcast / reply, bot info / quota / consumption, user profile, message content download
C. Webhook development	Local receiver (with signature verification), offline signature verification of a stored payload, replay to a local app, webhook endpoint get/set/test (repoint at a dev tunnel)
D. LIFF management	List / add / update / delete LIFF apps, update endpoint URL only (view.url)
E. Rich menu	Create / validate / list / get / delete, image upload & download, set / cancel default, link / unlink per user
F. Insight / Audience / Shop	Statistics (demographics, deliveries, followers, events, rich menu); audience group manage (incl. by-file user-id upload); mission sticker send

Target framework: net10.0.

---

Installation

Published on NuGet.org:

dotnet tool install -g Line.OpenApi.Tools

Run from local source (inside this repository):

dotnet run --project tools/Line.OpenApi.Tools -- <command> ...
# e.g.
dotnet run --project tools/Line.OpenApi.Tools -- --help

---

Quick start

1. Configure credentials (profiles)

Credentials are stored per profile in ~/.line/config.json (%USERPROFILE%\.line\config.json on Windows).

# Fastest start with a static token
line config set default --token "YOUR_CHANNEL_ACCESS_TOKEN"

# Or set the key material used for token issuance
line config set default \
  --channel-id "1234567890" \
  --kid "your-key-id" \
  --private-key "~/.line/keys/default.pem" \
  --secret "YOUR_CHANNEL_SECRET"

line config list          # list profiles (* marks the default)
line config get default   # show a profile (secrets masked)
line config use staging   # switch the default profile

Security: the config stores secrets in plain text. On Unix it is created with 0600 (owner read/write only). On Windows it inherits the user-profile ACL (a warning is printed on save). The private key is referenced by path only — its contents are never stored in the config.

2. Try it

line bot info                                   # bot information
line message push --to U0123... --text "Hello"  # push a message
line liff list                                  # list LIFF apps

---

Credential resolution order

Each command resolves credentials in the following priority (top wins):

1. Command-line arguments (--channel-token / --channel-id / --secret / --private-key / --kid)
2. Environment variables
3. Profile (--profile <name>, or the default profile)

Environment variables

Variable	Purpose
LINE_CHANNEL_ACCESS_TOKEN	Channel access token
LINE_CHANNEL_ID	Channel id
LINE_CHANNEL_SECRET	Channel secret (webhook signature verification, token revoke)
LINE_PRIVATE_KEY_PATH	Path to the RSA private key (PEM) for JWT assertions
LINE_KID	Key id of the signing key
LINE_PROFILE	Profile name to use
LINE_CONFIG	Override the config file path

Global options (all commands)

Option	Description
--profile <name>	Profile to use
--channel-token <t>	Token override (wins over env/profile)
--json	Emit machine-readable JSON
--verbose	Show details on error

---

Command reference

A. Token management — line token

# Issue (v2.1 by default; --kind stateless for a stateless token)
line token issue --kind v2.1 --days 30 \
  --channel-id 123 --kid KID --private-key ./key.pem

line token issue --store          # issue and store into the resolved profile
line token verify --token <t>     # check validity and remaining lifetime
line token revoke --token <t>     # revoke (requires --channel-id / --secret)

• Issuance requires the channel id, key id, and private key (PEM). The CLI builds and sends an RS256 JWT assertion.
• On the CLI the issued token is written to stdout and metadata to stderr (pipe-friendly). Use --store to save it into the resolved profile.

B. Message send & bot lookup — line message / line bot

Message content is provided via one of --text, --flex <file>, or --messages <file>.

line message push --to <id> --text "Hello"
line message push --to <id> --flex ./flex.json --alt-text "Notice"
line message push --to <id> --messages ./messages.json     # a messages-array JSON verbatim
line message multicast --to id1,id2,id3 --text "To many"
line message broadcast --text "To everyone"
line message reply --reply-token <token> --text "Reply"
line message content <messageId> -o ./image.jpg           # download a message's binary content

line bot info
line bot quota                # send quota limit
line bot quota consumption    # this month's consumption
line bot profile <userId>

• --flex takes a Flex message contents JSON and automatically wraps it into a Flex message with altText.
• --messages sends a [{ "type": "text", "text": "..." }, ...] messages array as-is.
• content is fetched from the data host (api-data.line.me); the facade routes it automatically.

C. Webhook development — line webhook

# Local receiver (verifies the signature and prints incoming events)
line webhook listen --port 5000

# Verify a stored payload's signature and summarize its events
line webhook verify --body ./payload.json --signature <x-line-signature>

# Replay a stored payload to a local app (no signature added; destination not validated)
line webhook replay --body ./payload.json --to http://localhost:5000/webhook

# Webhook endpoint configuration (channel access token; e.g. repoint at a fresh dev-tunnel URL)
line webhook get-endpoint                        # show the configured URL and active state
line webhook set-endpoint --url https://<tunnel>/callback
line webhook test-endpoint                       # ask LINE to probe the endpoint and report reachability

• Signature verification (listen / verify) requires the channel secret (profile / --secret); the endpoint commands (get/set/test-endpoint) use the channel access token instead.
• The URL for set-endpoint, test-endpoint, and liff update-url must be absolute https.
• Combine listen with an external tunnel (cloudflared / ngrok, etc.) to receive real webhooks from LINE. The tunnel itself is not bundled. set-endpoint then updates the LINE-side webhook URL without visiting the console.

D. LIFF management — line liff

line liff list                                # lists liffId + URL — use it to find the id
line liff add --file ./app.json               # add from a LIFF app definition JSON
line liff update <liffId> --file ./app.json   # full update from a JSON definition
line liff update-url <liffId> --url https://<tunnel>/   # update only view.url (https), e.g. a fresh dev tunnel
line liff delete <liffId>

E. Rich menu — line richmenu

line richmenu create --file ./richmenu.json    # create from a JSON definition; prints the new id
line richmenu validate --file ./richmenu.json   # validate a definition without creating it
line richmenu image <richMenuId> --file ./menu.png   # upload the image (PNG/JPEG; content type inferred)
line richmenu image-download <richMenuId> -o ./menu.png
line richmenu list
line richmenu get <richMenuId>
line richmenu delete <richMenuId>
line richmenu set-default <richMenuId>          # default for all users
line richmenu get-default
line richmenu cancel-default
line richmenu link <userId> <richMenuId>        # per-user link
line richmenu unlink <userId>
line richmenu id-of-user <userId>

• A typical dev cycle: create → image → set-default (or link to your own userId) → check on your device → iterate.
• Image upload requires PNG or JPEG; the content type is inferred from the file extension.

F. Insight / Audience / Shop — line insight / line audience / line shop

# Insight (statistics; all read-only; dates are yyyyMMdd)
line insight demographic
line insight deliveries 20260715
line insight followers 20260715
line insight events <requestId>
line insight per-unit <unit> --from 20260701 --to 20260715
line insight richmenu-summary <richMenuId> --from 20260701 --to 20260715
line insight richmenu-daily <richMenuId> --from 20260701 --to 20260715

# Manage Audience
line audience list --page 1 --size 20
line audience get <audienceGroupId>
line audience create --file ./create-audience.json        # with initial user IDs
line audience add-users --file ./add-audience.json         # carries audienceGroupId
line audience delete <audienceGroupId>
line audience upload-file --file ./user-ids.txt --description "my audience"   # one ID/IFA per line
line audience add-file <audienceGroupId> --file ./more-ids.txt

# Shop
line shop mission --file ./mission.json

• audience upload-file / add-file take a text file (one user ID or IFA per line) and are CLI-only — binary/file input is impractical over MCP.

Exit codes

Code	Meaning
0	Success
1	General error
2	Argument error (invalid input, missing input file, etc.)
3	Authentication / credential error
4	LINE API error (HTTP 4xx/5xx)

---

Use as an MCP server

line mcp starts an MCP server over stdio so AI agents can operate LINE as tools.

line mcp                       # all tools enabled
line mcp --read-only           # expose read-only tools only
line mcp --allow-secret-output # allow token issue to return the raw token (withheld by default)
line mcp --allow-remote-replay # allow non-loopback destinations for webhook replay (loopback-only by default)

Tool list

CLI commands are exposed as line_<area>_<verb> (except webhook listen). In the tables below, a ✓ in Read-only means the tool is available even under --read-only; ✗ means it mutates state and is excluded under --read-only.

General

Tool	Summary	Read-only
line_ping	Health check; returns "pong".	✓

Messaging

Tool	Summary	Read-only
line_message_schema	JSON Schema for message objects (build flex/template).	✓
line_message_push	Send a push message to a user/group/room.	✗
line_message_multicast	Send a message to multiple users.	✗
line_message_broadcast	Send a message to all friends of the bot.	✗
line_message_reply	Send a reply using a reply token.	✗

Flex Message preview

Tool	Summary	Read-only
line_flex_preview	Render Flex JSON in a live browser preview.	✓
line_flex_get_content	Read the JSON shown, including your browser edits.	✓
line_flex_validate	Structurally validate Flex JSON.	✓
line_flex_open	Reopen the preview tab.	✓

Bot

Tool	Summary	Read-only
line_bot_info	Bot info (userId, basicId, displayName, chat mode).	✓
line_bot_quota	Monthly message quota limit.	✓
line_bot_quota_consumption	This month's message consumption count.	✓
line_bot_profile	A user's profile by user id.	✓

Rich menu

Tool	Summary	Read-only
line_richmenu_schema	JSON Schema for a rich menu object.	✓
line_richmenu_list	List the channel's rich menus.	✓
line_richmenu_get	Get a rich menu by id.	✓
line_richmenu_get_default	Get the default rich menu id.	✓
line_richmenu_id_of_user	Get the rich menu linked to a user.	✓
line_richmenu_create	Create a rich menu (dryRun validates only).	✗
line_richmenu_delete	Delete a rich menu.	✗
line_richmenu_set_default	Set the default rich menu for all users.	✗
line_richmenu_cancel_default	Cancel the default rich menu.	✗
line_richmenu_link	Link a rich menu to a user.	✗
line_richmenu_unlink	Unlink the rich menu from a user.	✗

Image upload is CLI-only (line richmenu image <id> --file menu.png); binary input is impractical over MCP.

Insight

Tool	Summary	Read-only
line_insight_demographic	Friends' demographic attributes.	✓
line_insight_deliveries	Number of messages sent on a date.	✓
line_insight_followers	Number of followers as of a date.	✓
line_insight_events	Open/click stats of a message by request id.	✓
line_insight_per_unit	Stats for a custom aggregation unit over a period.	✓
line_insight_richmenu_summary	Aggregate rich-menu display/click stats.	✓
line_insight_richmenu_daily	Daily rich-menu display/click stats.	✓

Manage Audience

Tool	Summary	Read-only
line_audience_list	List audience groups (paginated).	✓
line_audience_get	Get an audience group and its jobs.	✓
line_audience_create	Create an audience group with initial user IDs.	✗
line_audience_add_users	Add user IDs to an audience group.	✗
line_audience_delete	Delete an audience group.	✗

By-file uploads (upload-file / add-file) are CLI-only; binary/file input is impractical over MCP.

Shop

Tool	Summary	Read-only
line_shop_mission	Send a mission sticker to a user.	✗

LIFF

Tool	Summary	Read-only
line_liff_list	List registered LIFF apps.	✓
line_liff_add	Add a LIFF app.	✗
line_liff_update	Update a LIFF app.	✗
line_liff_update_url	Update only a LIFF app's endpoint URL.	✗
line_liff_delete	Delete a LIFF app.	✗

Token

Tool	Summary	Read-only
line_token_verify	Verify a token's validity/lifetime (does not return the token).	✓
line_token_issue	Issue a token and store it in the profile.	✗
line_token_revoke	Revoke a token.	✗

Webhook

Tool	Summary	Read-only
line_webhook_verify	Verify a payload signature and summarize its events.	✓
line_webhook_get_endpoint	Get the configured webhook endpoint URL.	✓
line_webhook_test_endpoint	Ask LINE to send a test event; report reachability.	✓
line_webhook_replay	POST a payload to a local URL for debugging.	✗
line_webhook_set_endpoint	Set the channel's webhook endpoint URL.	✗

Rich menu dev cycle across MCP + CLI: assemble the menu with the agent (line_richmenu_schema → build JSON → line_richmenu_create with dryRun=true to validate, then create), then upload the image with the CLI (line richmenu image <id> --file menu.png) — binary upload is impractical over MCP, so it is intentionally CLI-only — and finally line_richmenu_set_default / line_richmenu_link and check on your device.

Each tool accepts an optional profile argument; credentials are resolved from the profile.

Flex Message preview (line_flex_*)

Preview a LINE Flex Message in a live, LINE-faithful browser view while you build it. The AI renders your JSON with line_flex_preview; a loopback web page opens and updates in place as you iterate. Adjust colors/spacing directly in the browser, then read them back with line_flex_get_content before sending. No LINE API calls or credentials are involved, so these tools are available under --read-only.

• line_flex_preview — render Flex JSON and open/update the preview → { ok, url, valid, warnings, opened }
• line_flex_get_content — get the JSON currently shown, including your browser edits → { content }
• line_flex_validate — structurally validate Flex JSON → { valid, warnings }
• line_flex_open — reopen the preview tab → { ok, url }

Env: LINE_FLEX_MCP_NO_OPEN (URL only, no auto-open), LINE_FLEX_MCP_STATE_DIR (state location), LINE_FLEX_MCP_ASSET_DIR (serve local images/video for preview — see below).

Previewing local images and video

LINE requires a public HTTPS URL for real media delivery, but while designing you often just want to see your own artwork. Set LINE_FLEX_MCP_ASSET_DIR to a folder (use an absolute path) and the preview server serves media files from it, so a Flex message can reference them by a relative url:

// with LINE_FLEX_MCP_ASSET_DIR=/path/to/flex-assets and flex-assets/assets/hero.png present
{ "type": "image", "url": "assets/hero.png", "size": "full" }

// a video component: the poster (previewUrl) is what renders in the preview; url is the mp4
{ "type": "video", "url": "assets/promo.mp4", "previewUrl": "assets/promo-poster.png",
  "altContent": { "type": "image", "url": "assets/promo-poster.png" }, "aspectRatio": "20:13" }

The served set matches what LINE renders in a Flex message: images are JPEG/PNG (APNG is .png) and video is .mp4 — other formats (GIF/WebP, etc.) are intentionally not served. The browser resolves assets/hero.png against the preview origin (http://127.0.0.1:<port>/). When you move the message to production, swap only the origin for your CDN/host — the relative path stays the same (https://cdn.example.com/assets/hero.png). Serving is opt-in (disabled unless the folder is set), confined to that folder (path traversal is rejected), loopback-only, and read-only-safe. This is a preview convenience: LINE itself renders neither local nor data: URLs.

The same browser renderer is also available as a Copilot App canvas extension (with a bundled zero-dependency Node MCP server as an alternative for Claude Desktop/Code) — see extensions/line-flex-viewer/.

Building messages with an AI agent (flex / template)

A primary MCP use case is a build → preview → adjust → send loop: the agent assembles a rich message, you see it rendered exactly like the LINE app, tweak it, and send once it looks right. Three tools make this reliable:

• line_message_schema(type) returns the JSON Schema for LINE message objects so the agent builds a shape-valid message. type is one of all / flex / template / imagemap / quickReply / action (default flex); it is read-only and returns no secrets. The schema is built from the same OpenAPI spec Kiota generates the client from, so it always matches the models. Types are emitted as named definitions (JSON Schema's $defs) that reference each other by pointer ($ref) instead of being expanded inline — necessary because a Flex box can contain other boxes (it is self-recursive), so inlining every reference would nest forever.
  • Simple messages (text / image / video / audio / location / sticker) are trivial and shown inline in the send-tool descriptions — you usually only need the schema for flex or template.
• line_flex_preview renders the Flex JSON in a live, LINE-faithful browser view (see the section above). Rather than guessing from raw JSON, you see the bubble/carousel, adjust colors and spacing right in the browser, and the agent reads your changes back with line_flex_get_content. No credentials involved.
• dryRun: true on the send tools (line_message_push / multicast / broadcast / reply) parses and shape-checks the messages and returns their parsed types without sending (no API call, no credentials required). A final safety check before the real send.

Typical flow: line_message_schema type=flex → build the Flex JSON → line_flex_preview (see it, tweak in the browser) → line_flex_get_content (pick up your edits) → line_message_push ... dryRun=true (validate) → line_message_push ... (send to your own userId).

Security design (MCP)

MCP tool results are assumed to enter the model's context (sent to the LLM provider, conversation history, logs), so the following protections are built in:

• line_token_issue does not return the raw token by default. The issued token is stored into the local profile and the result contains only metadata (tokenType / expiresInSeconds / keyId / maskedToken / storedProfile). Subsequent send tools operate via the stored profile. To obtain the raw token, start the server with --allow-secret-output and pass reveal: true in the tool call.
• line_webhook_replay allows loopback destinations only by default (SSRF mitigation). Remote destinations require --allow-remote-replay.
• Mutating / sending tools state their side effects in the description.

Register with Claude Code

claude mcp add line -- line mcp
# read-only registration
claude mcp add line -- line mcp --read-only

Register with Claude Desktop (claude_desktop_config.json)

{
  "mcpServers": {
    "line": {
      "command": "line",
      "args": ["mcp"]
    }
  }
}

---

In-process AI tools (Line.OpenApi.Extensions.AI)

The MCP server above lets an out-of-process agent (Claude Desktop / Claude Code) operate LINE. If instead you are building your own .NET agent, the companion package Line.OpenApi.Extensions.AI exposes the same Messaging operations as in-process Microsoft.Extensions.AI AIFunction tools — usable from Semantic Kernel or any Microsoft.Extensions.AI host, with no separate process. It depends only on Line.OpenApi.Messaging and Microsoft.Extensions.AI.Abstractions.

dotnet add package Line.OpenApi.Extensions.AI

using Line.OpenApi.Extensions.AI;
using Line.OpenApi.Messaging;

var line = MessagingClient.CreateWithStaticToken("CHANNEL_ACCESS_TOKEN");

// Safe by default: read-only tools only (bot info / quota / profile / message-validate).
IReadOnlyList<AIFunction> readTools = LineMessagingAiTools.CreateReadOnly(line);

// Sending is explicit opt-in and gated.
IReadOnlyList<AIFunction> tools = LineMessagingAiTools.Create(line, new LineAiToolOptions
{
    EnableSending  = true,                // enables push / multicast / reply (default false)
    AllowBroadcast = false,               // broadcast = largest blast radius, separate opt-in
    SendPolicy = (ctx, ct) =>             // bound blast radius (operation / recipients / count)
        new(ctx.Operation != LineSendOperation.Broadcast),
    BeforeSend = (ctx, ct) => /* human-in-the-loop / audit; inspect ctx.MessagesJson */ new(true),
});

// Give the tools to any Microsoft.Extensions.AI chat client:
//   new ChatOptions { Tools = [.. tools] }
// Or to Semantic Kernel:
//   kernel.Plugins.AddFromFunctions("Line", tools);

Tool names mirror the MCP tools (line_message_push, line_bot_profile, …). Safety gates (EnableSending / AllowBroadcast / SendPolicy / BeforeSend) are set by you at creation time and are never exposed as tool arguments, so a model cannot flip them. Sends are off by default; broadcast needs its own opt-in; results are non-secret. Rate / cumulative-count limiting is the host pipeline's responsibility, and message content / read-tool results flow to your LLM provider — treat tool arguments as potential PII in your logs.

Tools published by Line.OpenApi.Extensions.AI

The read/validate tools are always produced; the send tools are produced only when the matching LineAiToolOptions gate is set. The "Arguments" column is the whole argument list a model sees — the safety gates are not among them.

Tool	Arguments	Kind	Produced when
line_bot_info	(none)	read	Always
line_bot_quota	(none)	read	Always
line_bot_profile	userId	read	Always
line_message_validate	messagesJson	validate (never sends)	Always
line_message_push	to, messagesJson	send	EnableSending = true
line_message_multicast	to, messagesJson	send	EnableSending = true
line_message_reply	replyToken, messagesJson	send	EnableSending = true
line_message_broadcast	messagesJson	send (all friends)	EnableSending = true and AllowBroadcast = true

CreateReadOnly(client) produces exactly the four "Always" rows. DryRun = true does not change which tools are produced — the send tools are still present but only validate the payload and never contact the API (so SendPolicy / BeforeSend are not evaluated). SendPolicy / BeforeSend likewise never change the tool list; they can refuse an individual send at call time (LineSendRefusedException).

Released on its own cadence (tag ai-v*), separate from this CLI/MCP tool (tools-v*).

For a runnable end-to-end example (a scripted or real model driving the tools through the gates, offline by default), see samples/Line.OpenApi.Samples.Ai.

---

Build from source

dotnet build tools/Line.OpenApi.Tools/Line.OpenApi.Tools.csproj
dotnet test  tests/Line.OpenApi.Tools.Tests/Line.OpenApi.Tools.Tests.csproj

Related documentation

• Specification: docs/CLI-MCP-tool-spec.md
• Core libraries: repository root README.md

License

MIT (see LICENSE at the repository root).