The tool surface is small and the README does not pretend otherwise — it publishes an unchecked list of everything not yet built. That honesty is the reason to trust the part that does work: it reads dialogs and messages through your own API credentials, which makes "summarise what I missed overnight" a one-step job and makes it impossible for the assistant to send anything on your behalf.
A bridge between the Telegram API and an AI assistant. The README is explicit about the boundary: as of now the server provides read-only access. It signs in as you against the Telegram API, so what it can see is what your account can see.
- The list of dialogs — chats, channels and groups — `ListDialogs`
- The messages in a given dialog, with the option to restrict to unread ones
- A summary of what you missed, which is the use case the project calls out as working today
- The README's own roadmap of what is not built yet: marking a channel read, retrieving messages by date and time, downloading media, listing contacts and drafting messages are all listed unchecked
The `uv` tool, then `uv tool install git+https://github.com/sparfenyuk/mcp-telegram`. You need your own Telegram API ID and hash from my.telegram.org, and you sign in once with `mcp-telegram sign-in --api-id <your-api-id> --api-hash <your-api-hash> --phone-number <your-phone-number>`, entering the code Telegram sends — plus your password if two-factor is on. `TELEGRAM_API_ID` and `TELEGRAM_API_HASH` then go in the client config. `mcp-telegram logout` ends the session. The project asks you to read the Telegram API Terms of Service first, and notes that misuse of the API can get an account suspended.
One command plus a key — uv tool install git+https://github.com/sparfenyuk/mcp-telegram, then supply credentials