telegram-mcp
Enables use of a personal Telegram account within MCP clients for reading and sending messages, searching chats, and managing media, all running locally.
README
telegram-mcp
Your Telegram account in an MCP client. Signs in as you, not as a bot: it reads your private chats, and anything it sends comes from you.
Install
Node 22+.
-
Get API keys: my.telegram.org → API development tools → create an app (platform
Other, rest blank) → copy api_id and api_hash. -
Sign in — once, in a terminal:
TELEGRAM_API_ID=... TELEGRAM_API_HASH=... npx @vkruglikov/telegram-mcp loginIt asks for your phone, the code Telegram sends, and your 2FA password if you have one. This is a terminal command and not a tool on purpose — as a tool, the code and the password would pass through the model's context and land in chat logs.
-
Register the server:
claude mcp add telegram \ --env TELEGRAM_API_ID=... \ --env TELEGRAM_API_HASH=... \ -- npx -y @vkruglikov/telegram-mcpClients that read a JSON config instead:
{ "mcpServers": { "telegram": { "command": "npx", "args": ["-y", "@vkruglikov/telegram-mcp"], "env": { "TELEGRAM_API_ID": "...", "TELEGRAM_API_HASH": "..." } } } } -
Check:
npx @vkruglikov/telegram-mcp status.
logout revokes the session server-side and deletes it locally.
Features
Your account:
list_chats |
Dialogs, newest first, or only unread |
search_chats |
Find a dialog by name — Вадим also matches Vadim |
get_history |
Read a chat. Does not mark it read |
search_messages |
Full-text, one chat or the whole account |
send_message / send_file |
Delivered immediately, no undo |
download_media |
Save an attachment |
mark_read |
The only tool that marks anything read — the other side sees it |
list_contacts |
Saved contacts |
All of Telegram, not just your chats:
search_public_chats |
Channels, groups and people you have never opened |
search_public_posts |
Text of every public channel. A few free searches a day; the result says how many are left |
Prompts, shown as slash commands: personalize, draft_reply, catch_up.
Drafting in your voice
- Run
/personalize. It callssample_my_messagesto collect messages you sent in recent private chats, then describes the patterns: languages and how you mix them, message length, greetings and sign-offs, punctuation you skip, which emoji and where, recurring phrasings, how formality shifts, and what you never do. - It shows the profile before saving. Edit or reject it.
- Saved to
~/.telegram-mcp/style.md. Reconnect the server to load it. /draft_reply <chat> [intent]writes in that voice and shows the draft. It never sends.
get_style_profile prints the profile, forget_style_profile deletes it.
The profile covers how you write, not what you think. Facts, opinions and commitments stay yours to supply.
Everything stays on your machine
Local process, straight to Telegram, no service in between. Everything lives in
~/.telegram-mcp:
session.db— your login sessionstyle.md— your style profile, if you made onedownloads/— downloaded attachments
No telemetry, no account of ours, no third party. Credentials, session and messages go nowhere else.
Treat session.db like a password. It grants full access to your account without the
password and without 2FA. Never copy it to another machine or into a repo. If it leaks,
revoke it in Telegram → Settings → Devices. Deleting the folder by hand revokes nothing —
use logout.
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。
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。