An 80-tool Telegram bridge for Claude, Cursor, and other MCP hosts

What it does

This is an MCP server built on Telethon that maps a Telegram account into over 80 callable tools for compatible AI clients. It covers the full surface area of day-to-day Telegram usage: reading chats, sending or editing messages, managing groups and contacts, handling media, and even tweaking privacy settings. Results are sanitized and returned as structured JSON rather than raw Telegram blobs.

The interesting bit

The multi-account routing is unusually thoughtful: you can suffix environment variables like TELEGRAM_SESSION_STRING_WORK and the server will fan out read-only queries across identities while requiring an explicit account label for anything mutating. There is also a read-only tool-surface mode, though the README is honest that this is a registration restriction, not a Telegram session sandbox.

Key highlights

• 80+ tools covering chats, messages, contacts, media, folders, drafts, and admin operations
• Multi-account support with labeled session routing and per-account proxy overrides
• File-path security with allowed roots, path traversal rejection, and size/extension limits for media tools
• Read-only mode to prevent MCP clients from registering write-capable tools
• Docker support and proxy support for SOCKS4/5, HTTP, and MTProxy

Caveats

• The telegram-mcp name on PyPI belongs to an unrelated project; installing from PyPI could leak your API credentials to third-party code
• Read-only mode restricts which tools are exposed through MCP, but the underlying Telegram session still retains full account authority inside the server process
• File-path tools are disabled by default until allowed roots are explicitly configured

Verdict

Worth a look if you want your AI assistant to actually operate your Telegram account rather than just summarize screenshots. Skip it if you were hoping for a hosted, zero-setup integration—this requires your own API credentials, session strings, and a local runtime.