telegram-executor

💬 Telegram async executor for yaml-mcp-server: asks user to choose one of 2-5 options (or send custom text/voice).

🇷🇺 Русская версия: README_RU.md

What it does

telegram-executor receives async execute requests from yaml-mcp-server, sends a Telegram message with options, waits for user selection, and sends webhook callback back to yaml-mcp-server.

Supported flow:

• choose one predefined option (2..5)
• choose Custom option and reply with text or voice
• timeout handling with callback status=error

Request flow

1. yaml-mcp-server calls POST /execute and gets 202 Accepted.
2. telegram-executor sends a Telegram message with option buttons.
3. User clicks an option or sends custom text/voice.
4. telegram-executor sends callback to yaml-mcp-server webhook URL.

Installation

Requirements: Go >= 1.25.5.

go install github.com/codex-k8s/telegram-executor/cmd/telegram-executor@latest

Environment variables

All variables are prefixed with TG_EXECUTOR_:

• TG_EXECUTOR_TOKEN - Telegram bot token (required)
• TG_EXECUTOR_CHAT_ID - allowed Telegram chat id (required)
• TG_EXECUTOR_HTTP_HOST - HTTP listen host (required)
• TG_EXECUTOR_HTTP_PORT - HTTP listen port (default 8080)
• TG_EXECUTOR_LANG - message language (en/ru, default en)
• TG_EXECUTOR_EXECUTION_TIMEOUT - max wait time (default 1h)
• TG_EXECUTOR_TIMEOUT_MESSAGE - custom timeout note in Telegram (optional)
• TG_EXECUTOR_WEBHOOK_URL - Telegram webhook URL (optional)
• TG_EXECUTOR_WEBHOOK_SECRET - Telegram webhook secret (optional)
• TG_EXECUTOR_OPENAI_API_KEY - OpenAI API key for voice transcription (optional)
• TG_EXECUTOR_STT_MODEL - STT model (default gpt-4o-mini-transcribe)
• TG_EXECUTOR_STT_TIMEOUT - STT timeout (default 30s)
• TG_EXECUTOR_LOG_LEVEL - debug|info|warn|error
• TG_EXECUTOR_SHUTDOWN_TIMEOUT - graceful shutdown timeout (default 10s)

Webhook mode is enabled only when both TG_EXECUTOR_WEBHOOK_URL and TG_EXECUTOR_WEBHOOK_SECRET are set.

API

POST /execute

Request example:

{
  "correlation_id": "req-123",
  "tool": {
    "name": "telegram_request_feedback",
    "title": "Request user feedback"
  },
  "arguments": {
    "question": "Which rollout strategy should we apply?",
    "context": "Production deployment for billing-api",
    "options": [
      "Canary for 10% traffic",
      "Blue/green switch",
      "Delay rollout"
    ],
    "allow_custom": true
  },
  "spec": {
    "kind": "telegram_feedback_v1",
    "options_min": 2,
    "options_max": 5
  },
  "lang": "en",
  "markup": "markdown",
  "timeout_sec": 3600,
  "callback": {
    "url": "http://yaml-mcp-server.codex-system.svc.cluster.local/executors/webhook"
  }
}

Response:

{
  "status": "pending",
  "result": "queued",
  "correlation_id": "req-123"
}

Callback payload (to yaml-mcp-server)

Success example:

{
  "correlation_id": "req-123",
  "status": "success",
  "result": {
    "question": "Which rollout strategy should we apply?",
    "selected_option": "Canary for 10% traffic",
    "selected_index": 0,
    "custom": false,
    "input_mode": "button"
  },
  "tool": "telegram_request_feedback"
}

Custom voice/text example has custom=true and input_mode set to text or voice. Button label for custom option is fully controlled by telegram-executor i18n (TG_EXECUTOR_LANG or request lang).

Error example:

{
  "correlation_id": "req-123",
  "status": "error",
  "result": "execution timeout",
  "tool": "telegram_request_feedback"
}

Voice transcription

If TG_EXECUTOR_OPENAI_API_KEY is set, voice messages are transcribed via OpenAI.

ffmpeg is required:

sudo apt-get install -y ffmpeg

Security notes

• Service is stateless.
• Only one configured chat can interact with requests.
• Callback endpoint has no shared secret by default - protect it with network controls.

License

See LICENSE.