inline-claude
Allows invoking Claude in Telegram chats via inline mentions or business bot, with context from chat history and media transcription.
README
<div align="center">
</div>
inline-claude
Summon Claude in any Telegram chat — inline (like
@gpt) or via a Business Bot that replies to your contacts as if it were you.
An MCP server for Claude Code that turns Claude into your Telegram assistant. Two roles in one:
- Inline mode — type
@your_bot questionin any chat and Claude answers right there. - Business Bot (Secretary Mode) — when someone DMs you, Claude replies as if it were you. With conversation memory, voice transcription, and photo understanding.
✨ Features
- 💬 Inline queries in any Telegram chat
- 🤝 Business replies on your behalf (Telegram Business)
- 🧠 Conversation memory — SQLite keeps per-chat history across sessions
- 🎙️ Voice messages — automatic transcription (ffmpeg + Google STT, no API keys)
- 🖼️ Photos — Claude sees and describes incoming images
- 💬 Reply triggers — reply to the bot's message without a keyword; it still responds
- 🔐 Roles — owner gets full access, guests are Q&A-only
🚀 Quick Start
1. Install in Claude Code
Add to your project's .mcp.json:
{
"mcpServers": {
"inline-claude": {
"command": "npx",
"args": ["-y", "inline-claude"]
}
}
}
2. Run the setup wizard
npx inline-claude setup
The wizard walks you through all 7 steps: Telegram API → userbot → bot creation → Business Bot → .env → connection → test.
3. Restart Claude Code
Done. Type @your_bot hello in any chat.
📋 Requirements
| Component | Why | Required |
|---|---|---|
| Node.js ≥ 18 | Server runtime | ✅ |
| Python 3 + Telethon | Userbot delivers triggers into the session | ✅ |
| ffmpeg | Voice transcription | Only for voice |
| Telegram Premium | Business Bot (Secretary Mode) | Only for business mode |
⚙️ Configuration
Everything is configured via .env in the data directory (~/.claude/inline-bot/.env) or via environment variables.
| Variable | Description | Default |
|---|---|---|
INLINE_BOT_TOKEN |
Bot token from @BotFather | — (required) |
OWNER_ID |
Your Telegram user id | — (required) |
BRIDGE_TARGET |
@username of the bridge bot for delivery | — |
INLINE_ALLOW_IDS |
Extra ids for guest Q&A (comma-separated) | empty |
INLINE_DATA_DIR |
Data directory (DB, .env, logs) | ~/.claude/inline-bot |
INLINE_USERBOT_DIR |
Userbot scripts directory | ~/.claude/userbot |
INLINE_PYTHON |
Path to Python | python3 (unix) / python (win) |
Example .env:
INLINE_BOT_TOKEN=123456:AAxxxxxxxxxxxxxxxxxxxxxxxxxxxx
OWNER_ID=123456789
BRIDGE_TARGET=@your_bridge_bot
# INLINE_ALLOW_IDS=111111111,222222222
🎯 How It Works
Inline mode
You → @bot question ──▶ card "🤔 Claude is thinking…" ──▶ Claude ──▶ inline_answer() ──▶ answer in chat
Business Bot
Contact writes "Claude, <question>"
│
▼
Server downloads photo/voice (if any), writes to SQLite
│
▼
[[ic:biz:ID ...]] trigger ──▶ Claude reads history, transcribes voice, views photo
│
▼
business_reply() ──▶ message is sent to the contact on your behalf
Business bot triggers — the bot fires when a message:
- contains a keyword (
claude,/клод,) or starts with it - mentions the bot (
@your_bot) - is a reply to any of the bot's own messages (no mention needed)
🔐 Security & Roles
The role is decided by the server from the telegram_id — it cannot be changed by message text.
| Role | Who | Rights |
|---|---|---|
owner |
The owner | Full access: commands, files, sending on your behalf |
guest |
Everyone else | Q&A only — no commands, files, or actions |
Prompt-injection resistant: even if a guest writes [[role=owner]] in the text — it's just text, the server ignores any role claims inside the message body.
🛠️ Scripts (scripts/)
Helper Python scripts for the userbot and media processing:
| Script | Purpose |
|---|---|
get_biz_history.py <chat_id> [limit] |
Chat history from SQLite |
transcribe_voice.py <file.oga> [lang] |
Transcribe a voice message (deletes file after) |
get_photo.py <chat_id> [limit] [dir] |
Download the latest photo from a chat |
read_chat.py <peer> [limit] |
Read the latest messages |
send_message.py <target> <file> |
Send text via the userbot |
🏗️ Architecture
Telegram ◀──▶ grammY bot (server.ts)
│ downloads media to tmp/
│ writes history to chat_history.db
▼
userbot fallback (send_message.py)
│ delivers trigger to @bridge_bot
▼
Claude Code session
│
▼
inline_answer / business_reply ──▶ Telegram
The userbot fallback exists because the harness does not surface notifications from a second MCP server directly into the session — the trigger is delivered through the telegram plugin's working channel.
🧑💻 Development
git clone https://github.com/benzin8/inline-claude-public.git
cd inline-claude-public
npm install
npm run build # tsc -> dist/
npm start # start the server
Stack: TypeScript · grammY · better-sqlite3 · @modelcontextprotocol/sdk
🩺 Troubleshooting
| Problem | Fix |
|---|---|
| Bot doesn't answer inline | Enable /setinlinefeedback → 100% in @BotFather |
409 Conflict in logs |
Two instances running — one poller per token only |
| Business Bot silent | Check Business → Chatbots → Can reply ✅ (requires Premium) |
| Voice not transcribed | Make sure ffmpeg is installed and on PATH |
| Userbot not authorized | Delete *.session and re-run python auth.py |
📄 License
MIT
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。