TSGram MCP 🚀

Get Claude Code in your local project talking to Telegram in 3 minutes!

TSGram MCP connects Claude Code sessions to Telegram using TS/Node/Docker/cli-utils, enabling AI-powered code assistance directly in your Telegram chats. Ask questions about your codebase, get AI insights, and even edit files - all from Telegram on your phone!

🔒 How It Works (All Local)

Local Docker container runs the Telegram webhook server
TSGram MCP server runs locally
Your Claude Code or Claude Desktop runs locally and adds MCP locally
Create and delete bots in-app through Telegram, access limited to your telegram user id
Modern local web dashboard to manage bots from localhost
Local means local! If your computer is offline, so are your bots.

Quick Start (3 Minutes)

🤖 AI Agent Enhanced Setup (Recommended!)

Let Claude handle the entire setup and ask you for the required values.

From your terminal: 0. Install Docker Desktop and Node.js 20+ (npm included)

Clone the repository:

git clone https://github.com/areweai/tsgram-mcp.git
cd tsgram-mcp

From the command line, start Claude:
```
claude --model sonnet
```
Initialize with /init
Copy and paste this prompt:

First, explain to the human how to register a new Telegram bot with @BotFather and get their bot token (TELEGRAM_BOT_TOKEN), and how to get their Telegram user ID from @userinfobot (for AUTHORIZED_CHAT_ID). They can work on getting these credentials while you set up the system. Then help the user set up tsgram-mcp for claude code. Do everything from installing node modules to creating and deploying the local docker containers. Finally, when everything is deployed, remind the user to configure their bot token and authorized chat ID along wither other required .env variables.

Make sure you replace the TELEGRAM_BOT_TOKEN and AUTHORIZED_CHAT_ID placeholders with the actual values.

In order to retrieve those values, you will have to message the following Telegram bots and follow their instructions:

@BotFather
@userinfobot

The AI will then handle all the setup steps for you, including:

Installing dependencies
Configuring environment variables
Building Docker containers
Starting services
Guiding you through bot registration
Setting up the MCP server (and allow you to extend functionality locally)

CLI no-AI Alternative If Hitting LLM troubles: Setup Shell Script Run from your project root:

# One-line install (macOS/Linux)
curl -sSL https://raw.githubusercontent.com/areweai/tsgram-mcp/main/setup.sh | bash

# Or if you prefer to review first:
curl -sSL https://raw.githubusercontent.com/areweai/tsgram-mcp/main/setup.sh > setup.sh
chmod +x setup.sh
./setup.sh

Adding TSGram to an Existing Project:

If you already have a project and want to add TSGram as an MCP server:

From your existing project root:

git clone https://github.com/areweai/tsgram-mcp.git .tsgram-mcp
cd .tsgram-mcp

Use Claude to set it up:
```
claude model --sonnet
/init
```
Paste this prompt:

"First, explain to the human how to register a new Telegram bot with @BotFather and get their bot token (TELEGRAM_BOT_TOKEN) from @UserBotInfoBot, and how to get their Telegram user ID from @userinfobot (for AUTHORIZED_CHAT_ID). They can work on getting these credentials while you set up the system. Then help me add TSGram MCP to my existing project in the parent directory. Set up the Docker containers and MCP configuration so I can use Telegram to interact with my project files. The project root is one level up from here."

This will configure TSGram to work with your existing project structure (and allow you to extend functionality locally).

Test Your Bot! 🎉

IMPORTANT: You will need to message your bot first before it can message you.

Send your bot a message and ask it something about your local project!

TSGram file permissions and dangerzone mode

⚠️ Security Note: TSGram has basic safeguards to not list, preview, or serve .env files but it is still strongly recommended that you do not use TSGram for anything highly sensitive as third party servers are part of the transport layer.

📊 Web Dashboard

Monitor and manage your bots through the beautiful web dashboard at http://localhost:3000:

The dashboard provides:

Real-time bot status and activity monitoring
System health indicators (MCP server, AI model, API keys)
Bot management - create, test, and monitor Telegram bots
Activity logs - track messages, responses, and system events
Quick actions - send test messages and manage bot configurations
run from local - npm run dashboard to view (chrome recommended)

📋 You Will Need

Docker Desktop
Node.js 20+ (npm included)
Telegram on a phone
Claude Desktop or Claude Code (CLI)
An OpenRouter API key (We recommend creating a new API key with a $1.00 limit. Trust, but limit!)
Chrome (Optional but highly recommended for local management UI)

Manual Setup Steps

1. Get Your Bot Token & User ID

Create Bot: Message @BotFather on Telegram
Send /newbot and follow prompts
Copy your bot token (keep this secret!)
Get Your User ID (IMPORTANT): Message @userinfobot
- This returns your numeric user ID (e.g., 123456789)
- Use the user ID, NOT your username for security (usernames can be changed)

2. Configure Environment

Edit .env with your values:

TELEGRAM_BOT_TOKEN=your_bot_token_here
# CRITICAL: Use your numeric user ID from @userinfobot, not your username!
AUTHORIZED_CHAT_ID=123456789  # Your numeric user ID from step 1

# Choose ONE AI provider:
OPENROUTER_API_KEY=your_openrouter_key  # For Claude (recommended with $1 limit)
# OR
OPENAI_API_KEY=your_openai_key         # For GPT-4

Security Note: The bot now uses your numeric user ID instead of username for authorization. This is more secure because:

User IDs never change
Usernames can be changed by users
User IDs cannot be guessed or easily discovered

3. Start TSGram

npm install
npm run docker:build
npm run docker:start

🎯 Key Features

"Stop" and "Start" commands in TSGram bot chat provide convenient output control
Select subset of unix commands supported (:h ls or :h cat $FILE)
Opt-in edit mode for vibe coding and debugging on the go (:dangerzone to enable)
AI-Powered Responses: Uses Claude 3.5 Sonnet or GPT-4
Code Understanding: Reads and analyzes your entire codebase
Web Dashboard: Monitor bots at http://localhost:3000

💬 Bot Commands

Regular messages: Get AI responses about your code
:h - Show help and available commands
:h ls [path] - List files in directory
:h cat filename - View file contents
:dangerzone - Enable file editing (careful!)
:safetyzone - Disable file editing
stop - Pause bot responses
start - Resume bot responses

🛠️ Common Commands

# View logs
npm run docker:logs

# Check health
npm run docker:health

# Stop services
npm run docker:stop

# Rebuild after changes
npm run docker:rebuild

# Access dashboard (Chrome recommended)
npm run dashboard

📜 Useful Scripts

The project includes several helper scripts in the scripts/ directory:

Setup & Configuration

setup.sh - Automated install script (can be run via curl)
configure-mcp.sh - Configure MCP settings for Claude Desktop/Code
fix-permissions.sh - Fix file permissions if needed

Updates & Maintenance

update-system.sh - Update TSGram to latest version
update-ai-context.sh - Update AI context files

Testing & Debugging

test-api-keys.sh - Verify your API keys are working
test-docker-setup.sh - Test Docker configuration

Run any script with:

./scripts/script-name.sh
# or
npm run script-name  # for npm-wrapped scripts

🔧 Advanced Setup

Using with Claude Code

Add to your Claude Code MCP settings:

{
  "tsgram": {
    "command": "docker",
    "args": ["exec", "-i", "tsgram-mcp-workspace", "npx", "tsx", "src/mcp-server.ts"],
    "env": {
      "TELEGRAM_BOT_TOKEN": "your_bot_token_here",
      "AUTHORIZED_CHAT_ID": "123456789"
    }
  }
}

Note: Replace 123456789 with your actual numeric user ID from @userinfobot

MCP Compatibility

TSGram uses the standard Model Context Protocol (MCP) format and should be compatible with any MCP-supporting IDE or tool. Tested with:

✅ Claude Code CLI
✅ Claude Desktop

Should work with (standard MCP format):

Cursor IDE
GitHub Copilot
Other MCP-compatible editors

Use the same Docker configuration pattern shown above, adapting the syntax for your specific MCP client.

CLI-to-Telegram Bridge

Forward Claude Code CLI responses to Telegram:

# Setup global command (may prompt for sudo password to create npm global install /usr/local/bin/claude-tg)
npm run setup

# Use instead of 'claude'
claude-tg "analyze this codebase"

🚨 Troubleshooting

Bot not responding?

# Check if services are running
npm run docker:health

# View logs for errors
npm run docker:logs

Can't connect to bot?

Verify bot token is correct in .env
Check your user ID: Make sure AUTHORIZED_CHAT_ID is your numeric user ID (not username)
- Get it from @userinfobot if unsure
Make sure you messaged the bot first
If you get "authorization not configured" error, set AUTHORIZED_CHAT_ID in your .env file

File edits not working?

Type :dangerzone to enable (one-time)
Check logs for permission errors
Ensure Docker has file access

📚 Project Structure

tsgram/
├── src/
│   ├── telegram-bot-ai-powered.ts    # Main bot logic
│   ├── telegram-mcp-webhook-server.ts # MCP integration
│   └── models/                        # AI providers
├── docker-compose.tsgram-workspace.yml # Main Docker config
├── .env.example                       # Environment template
└── data/                             # Persistent storage

🤝 Contributing

PRs welcome! Please:

Fork the repository
Create a feature branch
Add tests if applicable
Submit a pull request

📄 License

MIT - Feel free to use in your own projects!

Need help? Open an issue on GitHub or message your bot with questions!