Telegram MCP Server
MCP server that acts as a gateway to Telegram, providing AI-optimized tools for messaging, search, and chat management via MTProto. Supports multi-user authentication with QR login and HTTP/stdio transports.
README
<img alt="Hero image" src="https://github.com/user-attachments/assets/635236f6-b776-41c7-b6e5-0dd14638ecc1" />
Telegram MCP Server — Model Context Protocol (MCP) gateway for Telegram. 8 context-efficient tools, multi-tenant, MTProto bridge.
Try the Demo
- Open https://tg-mcp.l1979.ru/setup
- Scan the QR code from Telegram mobile (Settings → Devices → Scan QR) — no phone typing, no OTP, no 2FA. Or enter your phone number as fallback.
- Copy your Bearer token from the success page
Then choose your path:
MCP Client (AI assistants)
- From the setup page, download the
mcp.jsonfile - Add the server to your AI client and ask: "send hello to my saved messages in telegram"
Direct API (curl)
- Run the command below (replace TOKEN with yours):
curl -X POST "https://tg-mcp.l1979.ru/mtproto-api/messages.SendMessage" \
-H "Authorization: Bearer TOKEN" \
-H "Content-Type: application/json" \
-d '{"params": {"peer": "me", "message": "Hello!"}}'
How It Works
This server sits between your AI agent and Telegram's API:
Your agent → MCP/HTTP → this server → MTProto → Telegram
What it does: Authenticates you with Telegram (QR or phone/bot token), exposes 8 AI-friendly tools instead of 80+ micro-APIs, and bridges raw MTProto for power users. Multi-tenant — one server, many users, isolated sessions.
Features
| Feature | Description |
|---|---|
| :building_construction: Dual Transport | Stdio for local MCP clients, HTTP for remote deploys (http-auth production, optional http-no-auth for dev) |
| :closed_lock_with_key: Multi-User Authentication | Shared http-auth server: one Bearer token per user, one Telegram account per MCP connection. QR login for instant auth — no phone/OTP/2FA. |
| :dart: AI-Optimized | 8 consolidated tools vs 80+ micro-tools — context-efficient design, LLM-friendly API, MCP ToolAnnotations |
| :globe_with_meridians: HTTP-MTProto Bridge | Direct curl access to any Telegram API method with entity resolution and safety guardrails |
| :shield: Session ACL | Opt-in per-principal limits on http-auth (ACL_ENABLED) — chat lanes, read_only, blocked_peers, allow_mtproto, ACL_DENY_UNLISTED_PRINCIPALS; see SECURITY.md |
| :tv: QR & Web Setup | Scan QR from Telegram mobile for instant auth (no phone/OTP/2FA) or use phone/code/2FA fallback — live at /setup |
| :label: One Agent, Multiple Accounts | Optional PREFIX_MCP_TOOLS_WITH_ACCOUNT — when one agent uses several MCP connections (same server, different tokens), prefixes tool names so they do not collide; not needed for standard multi-user hosting |
| :rocket: MTProto Proxy Support | Connect via MTProto proxy with automatic Fake TLS (EE prefix) and standard proxy detection |
| :card_file_box: Unified Session Management | Single configuration system for setup and server; per-token session files on shared multi-user hosts |
| :cloud: S3 Session Storage | Store sessions in S3-compatible object storage for ephemeral hosted deployments (Smithery, Fly.io, Railway) |
| :mag_right: Intelligent Search | Global & per-chat message search with multi-query support and intelligent deduplication |
| :mag: Unified Message API | Single get_messages tool for search, browse, read by IDs, and replies - 5 modes in one |
| :speech_balloon: Universal Replies | Get replies from channel posts, forum topics, or any message with one parameter |
| :busts_in_silhouette: Smart Contact Discovery | Search users, groups, channels with uniform entity schemas, forum detection, profile enrichment |
| :file_folder: Folder Filtering | Filter chats by dialog folder (archived, custom folders) with integer ID or name matching |
| :envelope: Advanced Messaging | Send, edit, reply, post to forum topics, formatting, file attachments, and phone number messaging |
| :paperclip: Secure File Handling | Rich media sharing with SSRF protection, size limits, album support, optional HTTP attachment streaming |
| :outbox_tray: Inline File Uploads | Data: URI (base64) file uploads in files param — work in all transport modes, filenames preserved, images sent as photos |
| :microphone: Voice Transcription | Automatic speech-to-text for Premium accounts with parallel processing and polling |
| :zap: High Performance | Async operations, parallel queries, and memory-conscious batching |
| :shield: Production Reliability | Auto-reconnect, configurable logging, comprehensive error handling |
Prerequisite: Install
uv— see docs if you don't have it. Or use Docker (see Installation Guide).
Quick Start
1. Install and authenticate
Quickest path (remote server): Open /setup → scan QR → copy token (see Try the Demo).
CLI path (local stdio): Run fast-mcp-telegram-setup once to create a Telegram session — then fast-mcp-telegram serves it:
uvx --from fast-mcp-telegram fast-mcp-telegram-setup \
--api-id="your_api_id" \
--api-hash="your_api_hash" \
--phone-number="+123456789"
Bot token alternative (no phone, no OTP):
Set BOT_API_TOKEN instead of --phone-number. See Installation Guide.
2. Configure MCP Client
stdio mode (local): Add to your MCP client config (e.g. claude_desktop_config.json) — stdio (standard input/output) is the default transport for local MCP clients:
{
"mcpServers": {
"telegram": {
"command": "uvx",
"args": ["fast-mcp-telegram"],
"env": {
"API_ID": "your_api_id",
"API_HASH": "your_api_hash"
}
}
}
}
http-auth mode (remote): Add to your MCP client config (e.g. claude_desktop_config.json):
{
"mcpServers": {
"telegram": {
"url": "https://tg-mcp.l1979.ru/mcp",
"headers": {
"Authorization": "Bearer YOUR_TOKEN"
}
}
}
}
Get your token by scanning the QR code on the setup page or see Installation Guide for deploying your own server.
3. Start Using
{"tool": "search_messages_globally", "params": {"query": "hello", "limit": 5}}
{"tool": "get_messages", "params": {"chat_id": "me", "limit": 10}}
{"tool": "send_message", "params": {"chat_id": "me", "message": "Hello!"}}
Deploy to Remote Server
Deploy your own MCP server on a VDS — see Installation Guide for step-by-step instructions.
Available Tools
| Tool | Purpose | Key Features |
|---|---|---|
search_messages_globally |
Search across all chats | Multi-term queries, date filtering, chat type filtering |
get_messages |
Unified message retrieval | Search/browse, read by IDs, get replies (posts/topics/messages), date filtering in all modes |
send_message |
Send new message | File attachments (URLs/local/data URIs), classic formatting (markdown/html), parse_mode=rich Rich Messages, reply to forum topics |
edit_message |
Edit existing message | Classic or parse_mode=rich formatting |
find_chats |
Find users/groups/channels | Multi-term search, contact discovery, folder filtering, username/phone lookup |
get_chat_info |
Get detailed profile info | Member counts, bio/about, online status, forum topics, enriched data |
send_message_to_phone |
Message phone numbers | Auto-contact management, optional cleanup, file support (URLs/data URIs), parse_mode=rich |
invoke_mtproto |
Direct Telegram API (power user) | Raw MTProto methods, entity resolution, safety guardrails — see MTProto Bridge |
See Tools Reference for detailed documentation with examples.
Documentation
- Installation Guide - Local setup and remote server deployment
- Tools Reference - Complete tools documentation
- MTProto Bridge - Direct API access via curl
- Contributing - Guidelines for contributors
- Security - Security features and best practices
Telemetry
Anonymous tool telemetry since v0.30.1 — heartbeat every 6h, no credentials or message content collected. Opt out with DO_NOT_TRACK=1. See ADR 0005.
Auth flow telemetry since v0.38.0 — atomic events during setup (phone, QR, bot token, reauthorize). Buffered flush on flow completion. See ADR 0008.
License
MIT License - see LICENSE
mcp-name: io.github.alexeyleshchenko/fast-mcp-telegram
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。
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。
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。
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。
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。
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。
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。
mcp-server-qdrant
这个仓库展示了如何为向量搜索引擎 Qdrant 创建一个 MCP (Managed Control Plane) 服务器的示例。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。