Open-source Telegram MCP server logs in as you via MTProto, read-only by default
TheVilfer · reddit · 2026-10-02
The author released better-tg-cli (MIT), an open-source Telegram CLI that also runs as an MCP server, letting Claude Code, Codex, Cursor, or Claude Desktop read the chats from your real account.
- Why not a bot: bots only see chats they've been added to and can't read prior history. This logs in as you over MTProto, the same protocol Telegram Desktop uses.
- Tools: telegramread handles inbox/unread, cross-chat search, reply threads, and forum topics — and can never write regardless of arguments. telegramwrite (send/edit/delete, inline buttons, file downloads) is off until a human runs telegram write-access on --for 1h via a terminal prompt or OS dialog the agent can't answer; writes only find chats by ID/@username/exact title and are all logged locally.
- Security: session stays in macOS Keychain / Linux Secret Service / Windows DPAPI or 1Password; QR login happens on a local 127.0.0.1 page so the agent never sees the code or 2FA password.
- Install: npx -y better-tg-cli mcp, Claude Code plugin marketplace, Homebrew, Scoop, MCP Registry.
The author is actively soliciting feedback on the permission model, noting prompt injection from random group chats as the real risk.
More from coding & agent
- NVIDIA paper: a better judge lifts terminal agent success from 50% to 68% without retraining — rohanpaul_ai · 2026-10-02
- Neuro-Symbolic Computer Use: agents that turn execution experience into self-healing policies, claimed 99% cheaper — xwang_lk · 2026-10-02
- Stripe now pays gas fees for agent stablecoin payments over MPP — jeff_weinstein · 2026-10-02
- The Flag Game: a toy setting to study agent swarm dynamics and cooperation — Hidenori8Tanaka · 2026-10-02
- Coinbase Link ships API to prove agents act on behalf of verified users — jeff_weinstein · 2026-10-02
- Dev deploys AI agent to 'stand guard' over open-source repo and flag needed changes — IanArawjo · 2026-10-02