tlgrm

An unofficial, feature-rich command-line client, MCP server, and webhook daemon for Telegram, built on Telethon.

Drive your personal Telegram account from the terminal — list chats, send and edit messages, pull history, download media, react, pin, schedule, run polls — and bridge incoming messages to an HTTP webhook in real time. Also ships as an MCP server so AI assistants like Claude can drive Telegram on your behalf.

Unofficial app notice: tlgrm is an independent, unofficial client built on the Telegram API (via Telethon). It is not affiliated with, endorsed by, or sponsored by Telegram.

Table of Contents

• Features

• How it works

• Installation

• Telegram API credentials

• Quick start

• Commands

• MCP server

• Claude Code skill

• Configuration

• Webhooks & background daemon

• Speech-to-text (optional)

• Output format

• Security

• Troubleshooting

• Documentation

• Contributing

• License

• Disclaimer

Related MCP server: mcp-telegram

Features

• 🔐 Personal account access — acts as you (user account, MTProto), not a bot, so it can read history and list members.

• 💬 Messaging — send text, files, media, and voice notes; reply, edit, forward, react, pin, mute, save to Saved Messages.

• 📜 History & members — fetch recent messages or participant lists from any chat.

• 📁 Chats — list your recent dialogs with unread counts and types.

• 🛠️ Group management — create groups/channels, add/remove members, leave chats.

• 📅 Scheduled messages & polls — schedule a text message at a future time, or send a poll/quiz.

• 🤖 MCP server — expose ~24 Telegram tools to any MCP-compatible AI assistant (Claude Desktop, Claude Code, etc.) with tiered read/write/destructive permissions.

• 🔔 Real-time webhooks — forward every incoming message to an HTTP endpoint as JSON, with automatic media download. Self-destructing (TTL) media is skipped; the webhook payload includes media.self_destruct: true.

• ⚙️ Background daemon — run the listener as a managed systemd user service.

• 🗣️ Optional speech-to-text — auto-transcribe incoming voice notes locally with Whisper.

• 🖥️ ~30 commands — every command prints structured JSON to stdout, making tlgrm easy to pipe into jq or other tools.

How it works

tlgrm uses the Telegram client API (MTProto) through Telethon, authenticating as your personal account with a login code (and 2FA if enabled). Your authenticated session is stored locally and reused, so you only log in once.

Because it acts as your user account, tlgrm can do things the Bot API cannot (read arbitrary history, list group members, send as yourself). This is also why it requires your own api_id/api_hash — see below.

Installation

Requirements: Python ≥ 3.10. FFmpeg is only needed for optional speech-to-text.

From PyPI

# CLI only
pip install tlgrm

# CLI + MCP server
pip install "tlgrm[mcp]"

# CLI + Whisper speech-to-text
pip install "tlgrm[stt]"

# Everything
pip install "tlgrm[all]"

With uv (recommended)

# CLI only
uv tool install tlgrm

# CLI + MCP server
uv tool install "tlgrm[mcp]"

From source (development)

git clone https://github.com/ali-commits/tlgrm.git
cd tlgrm
pip install -e .

To enable optional voice-note transcription, install the stt extra (see Speech-to-text):

pip install "tlgrm[stt]"     # or:  uv tool install "tlgrm[stt]"

Telegram API credentials

tlgrm does not ship with Telegram API credentials — you must supply your own. This protects you from a single shared api_id being rate-limited for everyone.

1. Go to my.telegram.org → API development tools.

2. Create an application (Platform: Desktop). We suggest a name like tlgrm — the title must not contain the word "Telegram".

3. Copy the api_id (a number) and api_hash (a hex string).

4. Export them before running tlgrm:

export TG_API_ID=1234567
export TG_API_HASH=your_api_hash_here

Add these to your ~/.bashrc / ~/.zshrc to persist them across sessions. See docs/configuration.md for all options.

Quick start

# 1. Authenticate once (interactive: phone → code → 2FA)
tlgrm login

# 2. List your recent chats
tlgrm chats --limit 10

# 3. Send a message (target = @username, numeric chat ID, or phone)
tlgrm send --target @username --text "Hello from the terminal!"

# 4. Read the last 5 messages of a chat
tlgrm history --target @username --limit 5

Commands

--target accepts a @username, a numeric chat ID, or a phone number. A purely numeric value is treated as a chat ID.

Read

Write

Tier 3 — Group management & advanced

Daemon & listener

📖 Full reference with every flag, examples, and output shapes: docs/commands.md.

MCP server

tlgrm ships a stdio MCP server (tlgrm-mcp) that exposes Telegram operations as tools to any MCP-compatible AI assistant.

Installation

pip install "tlgrm[mcp]"

The MCP server requires a prior tlgrm login — it reuses the same session file as the CLI.

Claude Desktop configuration

Add this to your Claude Desktop claude_desktop_config.json:

{
  "mcpServers": {
    "tlgrm": {
      "command": "uvx",
      "args": ["--from", "tlgrm[mcp]", "tlgrm-mcp"],
      "env": {
        "TG_API_ID": "...",
        "TG_API_HASH": "..."
      }
    }
  }
}

Permission tiers

The MCP server is read-only by default. Write and destructive operations must be explicitly enabled at startup:

To enable write access in Claude Desktop, pass the flags via args:

"args": ["--from", "tlgrm[mcp]", "tlgrm-mcp", "--allow-write"]

Safety note: The read-only default ensures an AI assistant can never send or delete messages unless you have explicitly opted in. Always review what permissions you grant.

Claude Code skill

A Claude Code skill lives at skills/tlgrm/ in this repository. It teaches Claude Code how to drive tlgrm — both via the CLI and the MCP server — and includes a capability map, safety rules, and workflow recipes.

Installation

cp -r skills/tlgrm ~/.claude/skills/

Once installed, Claude Code will automatically use it when you ask things like "check my Telegram", "send a message to @someone", or "summarize my unread chats".

Configuration

All configuration is via environment variables. None are required except the API credentials.

See docs/configuration.md for the full reference including all STT backend options.

Webhooks & background daemon

tlgrm can forward every incoming message to an HTTP endpoint as JSON, downloading any attached media automatically:

tlgrm listen --webhook-url https://your-server.com/webhook \
             --webhook-header "Authorization: Bearer YOUR_TOKEN"

To run it persistently as a systemd user service:

tlgrm daemon install --webhook-url https://your-server.com/webhook
tlgrm daemon status
tlgrm daemon logs
tlgrm daemon uninstall

Note: self-destructing (TTL) media is not downloaded — the webhook payload includes "media.self_destruct": true but the file is skipped.

📖 Full guide, payload schema, and security notes: docs/webhook-guide.md.

Speech-to-text (optional)

When the stt extra is installed, the webhook daemon automatically transcribes incoming voice notes and audio, populating media.transcription in the webhook payload. You can also transcribe any audio file directly with tlgrm transcribe (no login required). tlgrm supports pluggable STT backends: local models and cloud APIs.

Default: faster-whisper (local, lightweight, auto-downloads the model on first use; CPU by default, set TG_STT_DEVICE=cuda for GPU).

tlgrm transcribe --file voice.ogg                       # default backend
tlgrm transcribe --file voice.ogg --backend faster-whisper

# Local faster-whisper (default — recommended)
pip install "tlgrm[stt]"     # requires FFmpeg on your system

# Original OpenAI Whisper
pip install "tlgrm[stt-whisper]"

# Cloud backends — just set an API key (no extra package needed)
export OPENAI_API_KEY=sk-...   # uses openai backend automatically
export GROQ_API_KEY=gsk_...    # uses groq backend automatically

If no STT backend is available, transcription is silently skipped — all other features work unchanged.

See docs/configuration.md for the full list of backends, selection precedence, config file format, and privacy considerations for cloud backends.

Output format

Every command prints pretty-printed JSON to stdout (non-ASCII preserved), making tlgrm easy to pipe into jq or other tools:

tlgrm chats --limit 3 | jq '.[].name'

Errors are reported as {"success": false, "error": "..."} with a non-zero exit code on authentication failure.

Security

• Your session file (~/.tlgrm/tg_session.session) grants full access to your Telegram account. Keep it private; it is git-ignored by default.

• The systemd unit file is written with owner-only permissions (0600) because it may embed webhook auth headers.

• Enable Two-Step Verification on your Telegram account.

• Found a vulnerability? See CONTRIBUTING.md.

Troubleshooting

Documentation

• Getting Started — install, log in, send your first message

• Configuration — environment variables and paths

• Command Reference — every command, flag, and output shape

• Webhook & Daemon Guide — webhooks, systemd, payload schema

Contributing

Contributions are welcome! Please read CONTRIBUTING.md for development setup, coding conventions, and how to report issues.

License

Released under the MIT License © 2026 Ali Alrabeei.

Disclaimer

tlgrm is an unofficial client and is not affiliated with, endorsed by, or sponsored by Telegram. Use at your own risk.