ccremote

Control Claude Code from your phone via Telegram.

ccremote bridges Claude Code CLI to Telegram DMs. Send prompts from your phone, get streaming responses via live draft previews, send photos/documents/voice messages — all without sitting at your computer.

How it works

Phone (Telegram)            Local Machine
──────────────────          ──────────────────
  DM with bot               ccremote
  "fix the bug"  ────►       ├─ Telegram bot (aiogram)
  ◄ streaming draft...       ├─ claude --print --resume <id>
  ◄ final response           └─ Whisper transcription

1. Run uvx ccremote . in any project directory
2. Chat with Claude in your bot's DM
3. Responses stream as live draft previews, then appear as final messages
4. Send photos, documents, or voice messages
5. Permission denials show inline buttons to approve and retry

Quick start

1. Create a Telegram bot — open @BotFather, send /newbot, follow the prompts, copy the token
2. Get your Telegram user ID — open @userinfobot, send /start, copy the number
3. Create a .ccremote file in your project directory:

   CCREMOTE_BOT_TOKEN=123456:ABC-DEF1234ghIkl-zyx57W2v1u123ew11
   CCREMOTE_ALLOWED_USER=123456789

4. Run:

   uvx ccremote .

5. Open your bot's DM in Telegram and start chatting with Claude

Tip: Add .ccremote to your global .gitignore — it contains secrets.

Prerequisites

• Claude Code CLI installed and authenticated
• Python 3.11+ and uv (recommended) or pip
• A Telegram bot token (see step 1 above)
• OpenAI API key (optional, only needed for voice message transcription)

Install locally (for development)

git clone https://github.com/nurikk/ccremote.git
cd ccremote
uv venv .venv
source .venv/bin/activate
uv pip install -e .

Configuration

Environment variables vs .ccremote file

You can configure ccremote in two ways — they can be combined:

Option A: Global environment variables (apply to all projects)

# Add to ~/.zshrc or ~/.bashrc
export CCREMOTE_BOT_TOKEN=123456:ABC-DEF1234ghIkl-zyx57W2v1u123ew11
export CCREMOTE_ALLOWED_USER=123456789
export CCREMOTE_OPENAI_API_KEY=sk-...   # optional, for voice messages

Option B: Per-project .ccremote file (overrides env vars for that project)

CCREMOTE_BOT_TOKEN=999888:XYZ-different-bot-token
CCREMOTE_ALLOWED_USER=123456789
CCREMOTE_OPENAI_API_KEY=sk-...

Priority: .ccremote file > environment variables > defaults.

Configuration options

Variable	Required	Default	Description
CCREMOTE_BOT_TOKEN	yes	—	Telegram bot token from @BotFather
CCREMOTE_ALLOWED_USER	yes	—	Your Telegram user ID
CCREMOTE_OPENAI_API_KEY	no	""	OpenAI key for voice transcription
CCREMOTE_LOG_LEVEL	no	info	debug, info, warning, error
CCREMOTE_INCLUDE_PARTIAL_MESSAGES	no	true	Include partial messages in stream
CCREMOTE_DRAFT_THROTTLE_MS	no	300	Min ms between draft updates
CCREMOTE_MAX_MESSAGE_LENGTH	no	4000	Max chars per Telegram message
CCREMOTE_CLAUDE_ALLOWED_TOOLS	no	—	JSON array of allowed tools (all if unset)

Usage

# With uvx (no install needed)
uvx ccremote .

# Or if installed locally
ccremote .
ccremote ~/code/myproject

That's it. The bot connects to Telegram and you can start chatting.

Message types

• Text — sent directly as prompts to Claude
• Photos — downloaded to .ccremote-attachments/ in the project, path passed to Claude
• Documents — same as photos, keeps original filename
• Voice messages — transcribed via OpenAI Whisper, sent as text (requires CCREMOTE_OPENAI_API_KEY)

Permission handling

When Claude tries to use a tool that's blocked by permissions, you'll see an inline keyboard:

⚠️ Permission denied:
  • Bash: ls ~/Downloads

[✅ Allow]  [❌ Skip]

Tapping Allow re-runs the prompt with the denied tools added to --allowedTools.

Session continuity

ccremote uses claude --print --resume <session_id> to maintain conversation context. Each message continues the same Claude session, so context builds up naturally across your conversation.

Architecture

src/ccremote/
├── cli.py          Entry point — starts bot, creates session
├── bot.py          aiogram dispatcher, message sending helpers
├── relay.py        Claude ↔ Telegram relay, streaming, permissions
├── markdown.py     Markdown → Telegram HTML converter
├── config.py       pydantic-settings configuration
└── models.py       Pydantic data models (Session)

Key design decisions:

• Single session mode — one ccremote process per project, DM-only. This is intentional: sendMessageDraft only works in private chats, supergroup forums don't allow bots to create new threads via the Bot API, making group-based workflows impractical
• Each prompt spawns claude --print --resume <id> (stateless process, persistent session)
• sendMessageDraft (Bot API 9.3+) for flicker-free live streaming
• Markdown converted to Telegram HTML for formatted output
• Per-project .ccremote overrides global env vars

Development

# Install dev dependencies
uv pip install -e ".[dev]"

# Run tests
python -m pytest tests/ -v

# Lint
ruff check src/ tests/

# Format
ruff format src/ tests/

# Type check
ty check src/

User Feedback

License

MIT