Telegram CLI

Use the local skill script for Telegram work on the user's personal account.

This skill exists because Telegram Bot API is the wrong tool for reading a real personal account. Use MTProto via Telethon instead.

Quick rules

Prefer reads first, then propose the action queue.
Write commands are dry-run by default and require --execute.
Never run any write command with --execute unless the user explicitly approved that specific action or batch first.
For send, always present a draft message first and ask the user for confirmation before sending.
Do not run send --execute unless the user explicitly approved the final recipient and text.
Mark-read/archive/mute are still Telegram writes; use them only after the user has approved the batch/action.
Do not add edit/delete/bulk export/background automation unless the user explicitly asks.
Treat the Telethon session like a high-privilege secret.
Assume unread preservation is best-effort until tested on a real chat.

Local setup

Prefer the skill-local script and cached virtualenv over any global CLI install. Prefer saved Telegram config over shell-exported environment variables once setup is complete. Treat the virtualenv under ~/.cache/telegram-cli/venv as generated local state, not part of the skill itself. If the installer drops skill-local dotfiles, the bootstrap script recreates .gitignore automatically.

Bootstrap the local environment:

<skill-path>/scripts/bootstrap_venv.sh

After bootstrap, use:

<skill-path>/scripts/telegram-cli

scripts/telegram-readonly remains as a backwards-compatible alias for older workflows.

If the cached virtualenv is missing later, just run the bootstrap script again.

Primary config path:

~/.config/telegram-cli/config.json

Recommended one-time setup:

Make sure api_id and api_hash are available.
Save them with:

<skill-path>/scripts/setup-api-key.sh

Run:

<skill-path>/scripts/telegram-cli auth

After successful login, the config file stores api_id, api_hash, and the Telegram session string so future reads do not need exported shell variables.

Commands

Show built-in help

<skill-path>/scripts/telegram-cli help

Authenticate once

<skill-path>/scripts/setup-api-key.sh
<skill-path>/scripts/telegram-cli auth

List chats

dialogs --query does token-based matching across name, username, and title, so queries like petros skynet work even when the exact full string is not present as one substring.

<skill-path>/scripts/telegram-cli dialogs --limit 50

Read recent messages

<skill-path>/scripts/telegram-cli messages --chat '@username' --limit 50 --reverse

Search messages

<skill-path>/scripts/telegram-cli search 'invoice' --limit 50

Restrict search to one chat:

<skill-path>/scripts/telegram-cli search 'deadline' --chat '@username' --limit 50

List recent unread chats

Default behavior is opinionated: exclude muted and archived chats.

<skill-path>/scripts/telegram-cli unread-dialogs --limit 10

Include muted and/or archived when needed:

<skill-path>/scripts/telegram-cli unread-dialogs --limit 10 --include-muted --include-archived

List recent unread DMs only

<skill-path>/scripts/telegram-cli unread-dms --limit 10

Send a message

Draft first in chat, ask the user to confirm, then dry-run:

<skill-path>/scripts/telegram-cli send --chat '@username' --text 'Thanks, will check.'

Send only after the user approves final text and recipient:

<skill-path>/scripts/telegram-cli send --chat '@username' --text 'Thanks, will check.' --execute

Mark read

<skill-path>/scripts/telegram-cli mark-read --chat 123456789
<skill-path>/scripts/telegram-cli mark-read --chat 123456789 --execute

Archive or unarchive

<skill-path>/scripts/telegram-cli archive --chat 123456789
<skill-path>/scripts/telegram-cli archive --chat 123456789 --execute
<skill-path>/scripts/telegram-cli archive --chat 123456789 --unarchive --execute

Mute or unmute

<skill-path>/scripts/telegram-cli mute --chat 123456789 --hours 8
<skill-path>/scripts/telegram-cli mute --chat 123456789 --hours 8 --execute
<skill-path>/scripts/telegram-cli mute --chat 123456789 --unmute --execute

Workflow

Read references/setup-and-safety.md if setup, auth, or unread-state behavior matters.
Ensure the cached virtualenv is bootstrapped.
Ensure Telegram API credentials exist.
Run auth once to create the session and write ~/.config/telegram-cli/config.json.
Use dialogs, messages, search, unread-dialogs, or unread-dms as needed.
For writes, get the user's approval first, run the dry-run, check the JSON target/action, then use --execute.
Keep usage narrow and intentional.

Expected outputs

The wrapper returns JSON. Parse it instead of relying on fragile text scraping.

Dialog objects include:

is_user
is_group
is_channel
is_bot
archived
muted
unread counters

Files

Launcher: scripts/telegram-cli
Launcher: scripts/telegram-readonly
Python implementation: scripts/telegram_cli.py
Local bootstrap: scripts/bootstrap_venv.sh
Credential setup helper: scripts/setup-api-key.sh
Setup notes: references/setup-and-safety.md
Config storage: ~/.config/telegram-cli/config.json
.env is optional fallback only; it is not the preferred long-term setup.
~/.cache/telegram-cli/venv is generated local state and can be recreated with <skill-path>/scripts/bootstrap_venv.sh.
ChatGPT/Codex MCP server: mcp/server.mjs

ChatGPT/Codex MCP

The MCP server exposes typed read tools for dialogs, messages, search, and unread lists. Telegram writes use a mandatory two-step flow: a telegram_prepare_* tool returns the resolved dry-run preview and a short-lived one-time token; telegram_execute_prepared_action can consume that frozen token only after the user explicitly approves the exact preview in a new message.

The MCP intentionally does not expose arbitrary shell arguments, local text-file paths, interactive auth, edit/delete operations, or background watchers. Run interactive Telegram authentication through the local CLI before starting the MCP.

Install and start locally:

cd <skill-path>/mcp
npm install
node server.mjs

The durable launcher is scripts/telegram-mcp. For an OpenAI Secure MCP Tunnel on this host, install the launcher and service template, create a separate tunnel and restricted runtime key, then run scripts/finish-telegram-mcp-tunnel.

When to stop and ask

Stop and ask before:

sending a Telegram message
enabling any background watcher/daemon
broad exporting of large chat histories
changing how secrets/session storage works

Docs

Fast lookup:

Telethon client reference: https://docs.telethon.dev/en/stable/quick-references/client-reference.html
Telethon TelegramClient API: https://docs.telethon.dev/en/stable/modules/client.html
Telegram folders/archive API: https://core.telegram.org/api/folders
Telegram notification settings API: https://core.telegram.org/method/account.updateNotifySettings