HN 표시: SciTeX 알림 – AI 에이전트에게 음성 제공: TTS, 전화 통화, SMS

Multi-backend notification system — give your AI agents a voice Full Documentation · pip install scitex-notification Developers delegate tasks to AI coding agents — and then wait. Staring at terminals wastes time and drains cognitive resources. Sitting for hours waiting for results takes a physical toll, too. SciTeX Notification lets you leave your desk — or even sleep — while your agents keep working. One MCP server gives them a voice in the following channels: TTS, phone calls, SMS, email, and webhooks. Supported Backends | Backend | API | Transport | Cost | Internet | Notes | |---|---|---|---|---|---| | Audio | alert() | TTS to local speakers | Free | No | Via scitex-audio; SSH relay supported | | Desktop | alert() | notify-send / PowerShell | Free | No | Linux / WSL | | Emacs | alert() | emacsclient message | Free | No | Popup, minibuffer, or alert | | Matplotlib | alert() | Visual popup window | Free | No | | | Playwright | alert() | Browser popup | Free | No | | alert() | SMTP | Free | Required | Gmail, SMTP relay | | | Webhook | alert() | HTTP POST | Free | Required | Slack, Discord, custom endpoints | | Twilio | call() sms() | Phone call / SMS | Paid | Required | Twilio setup needed | alert() tries backends in fallback order until one succeeds. call() and sms() target Twilio directly. When an AI agent needs your attention — even while you sleep — it can escalate from audio to phone call. Fig. 1: Left — Claude Code terminal showing audio → phone call escalation after 7 consecutive auditory feedback but with no response from the user. Right — iPhone receiving repeated "SciTeX Alert" calls from the AI agent. Penetrating iPhone Silent Mode: - Emergency Bypass (most reliable): Save your Twilio number as a contact → Ringtone → enable Emergency Bypass. All calls ring regardless of Focus/Silent mode. - Repeated Calls (fallback): iOS allows the second call from the same number within 3 minutes to ring through. repeat=2 triggers this automatically with a 30-second gap. Example Command for Claude Code (`/speak-and-continue`) The /speak-and-continue command is a real-world example of scitex-notification in action. It configures an AI coding agent (Claude Code) to: - Provide TTS feedback via audio_speak MCP tool while working autonomously - Repeat the same audio message when the user sends /speak-and-continue again (they may be away) - Count consecutive speaks without user response (spoke counter) - Escalate to phone call after 7 unanswered speaks — waking the user if asleep This enables a 24/7 development workflow: the agent works autonomously, speaks progress aloud, and calls your phone when it needs your attention or when all tasks are complete. Requires Python >= 3.10. pip install scitex-notification Install with optional backends: pip install "scitex-notification[audio]" # audio alerts via scitex-audio pip install "scitex-notification[twilio]" # SMS via Twilio pip install "scitex-notification[mcp]" # MCP server for AI agents pip install "scitex-notification[all]" # everything Python API import scitex_notification as notification notification.alert("Job done") # default backend (with fallback) notification.alert("Error", backend="email") # specific backend notification.alert("Critical", backend=["sms", "email"]) # multiple backends notification.call("Urgent!") # phone call via Twilio notification.sms("Build done!") # SMS via Twilio notification.available_backends() # list installed backends CLI Commands scitex-notification --help # Show all commands scitex-notification send "Job done" # Send via default backend scitex-notification send "Done" --backend email # Send via specific backend scitex-notification config # Show configuration scitex-notification backends # List available backends scitex-notification mcp list-tools # List MCP tools MCP Server — for AI Agents AI agents can send notifications and check system status autonomously. | Tool | Description | |---|---| notify_handler | Send an alert via specified or fallback backends | notify_by_level_handler | Send alert using level-based backend config | list_backends_handler | List all backends with availability status | get_config_handler | Show current notification configuration | scitex-notification mcp start Environmental Variables List .env.example . For AI Agents such as Claude Code. Example Add .mcp.json to your project root. Use SCITEX_NOTIFICATION_ENV_SRC to load all configuration from a .src file — this keeps .mcp.json static across environments: { "mcpServers": { "scitex-notification": { "command": "scitex-notification", "args": ["mcp", "start"], "env": { "SCITEX_NOTIFICATION_DEFAULT_BACKEND": "audio", "SCITEX_NOTIFICATION_TWILIO_SID": "ACxxxxxxx", "SCITEX_NOTIFICATION_TWILIO_TOKEN": "...", "SCITEX_NOTIFICATION_TWILIO_TO": "+XX-XXX-XXX-XXXX", "SCITEX_AUDIO_RELAY_PORT": "${SCITEX_AUDIO_RELAY_PORT}" } } } } Remote machines can speak from your local speakers. Setup example can be seen at ./docs/audio-relay-setup.src Example 1. Local machine (has spea