Skip to main content
Glama
line-relay

line-relay

by cmbaldwin

line-relay

Local stdio MCP server that drives the LINE macOS app (personal account) via the Accessibility API. No network listener. Sending needs prepare_message -> send_prepared(confirm_id); there is no direct send.

⚠️ LINE terms of use: read before using

LINE offers no API for personal accounts, and its terms restrict automated or unauthorized use of the service. A LINE account can be restricted or banned for it. line-relay is meant only as a personal assistive tool that stays inside those limits as closely as possible:

  • Your own account, on your own Mac. The LINE desktop app must already be logged in. It never handles credentials and never talks to LINE's servers directly.

  • A human approves every message. Each send needs a prepared draft and an explicit confirm. There is no unattended or scheduled sending, and no direct-send tool.

  • Low volume. Hard limits of 20 sends per hour and 1 per 10 s. It is for the few messages a day you would otherwise type yourself.

  • Not for bulk messaging, marketing, spam, scraping, unsolicited contact, bots that answer automatically, or any commercial or business messaging. For those, use a LINE Official Account and the Messaging API, which is LINE's supported way to automate.

You are responsible for following LINE's current terms of use. Use at your own risk. This project is not affiliated with or endorsed by LINE / LY Corporation.

Related MCP server: line-desktop-mcp

Setup

cd ~/Dev/line-relay && uv venv --python 3.13 .venv && uv pip install --python .venv/bin/python -r requirements.txt
.venv/bin/python test_offline.py

Permissions (System Settings > Privacy & Security)

The process that launches the server needs both:

  • Accessibility (read LINE's UI, press buttons, send keys)

  • Screen Recording (screenshots of the composed message)

When launched by Claude desktop, that is the Claude app (child processes inherit its grants); the interpreter is ~/Dev/line-relay/.venv/bin/python. If a tool reports empty windows or blank screenshots, grant the permission and restart Claude.

Claude desktop config (~/Library/Application Support/Claude/claude_desktop_config.json)

"mcpServers": { "line-relay": {
  "command": "/path/to/line-relay/.venv/bin/python",
  "args": ["/path/to/line-relay/server.py"] } }

Behaviour notes

  • LINE must be on the current Space: its AX tree is empty otherwise. Each tool call activates LINE, then restores the previous app and the previously open chat. See docs/line-ax-notes.md.

  • Limits: 20 sends/hour, 1 per 10 s. confirm_id expires after 5 min. Log: ~/Library/Logs/line-relay/log.jsonl.

  • send_prepared refuses if the focused chat or composer text no longer matches.

Test checklist

  • test_offline.py passes (bad/expired confirm_id).

  • find_chat("テストグループ") returns the chat.

  • Text to Keep/test chat: prepare -> screenshot -> send_prepared.

  • Image and PDF to Keep/test chat (verify attach flow, see notes).

  • Wrong/expired confirm_id sends nothing; log records rejection.

  • LINE quit: status() and prepare_message() return clear errors.

  • After each call the previous app/chat is restored.

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    Related MCP Connectors

    Related MCP Servers

    • A
      license
      B
      quality
      D
      maintenance
      Enables AI assistants to send LINE messages, including text, stamps, Flex Messages, and check message quotas through the LINE Messaging API.
      1
      MIT
    • A
      license
      Not graded
      quality
      D
      maintenance
      Enables AI tools to read and send messages through LINE Desktop via MCP, supporting manual or automatic sending without official LINE API tokens.
      151 npm
      130
      MIT
    • A
      license
      Not graded
      quality
      D
      maintenance
      Enables AI tools to read and send LINE messages by automating the LINE Desktop application, with macOS OCR-based reading and idle-aware scheduling.
      1
      MIT
    • F
      license
      Not graded
      quality
      C
      maintenance
      Enables reading messages from explicitly permitted macOS KakaoTalk chat rooms and sending replies through a two-step prepare/commit approval flow via the accessibility API. Designed fail-closed, with sending and automation disabled by default.
      -