tonydzi

Telegram MCP Kit — Development skill for Claude Code

Development community

Connect Claude (or any MCP client) to your own Telegram in ~15 minutes: setup prompt for Claude Code/Codex, production patches (shared daemon, extra tools, multi-account), watchdog, and every gotcha t.

How to install Telegram MCP Kit

This entry records only its repository, not the path inside it, so there is no exact command to give. Open tonydzi/telegram-mcp-kit and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What Telegram MCP Kit does

Alternatives in Development

Agentlas Connect — Connect Agentlas agents or teams to Telegram 1.1k ★
Openclaw Billing Proxy — Route OpenClaw API requests through Claude Code subscription billing instead of Extra Usage 460 ★
Clauditor — Stop Claude Code from burning through your quota in 20 minutes 367 ★

README

telegram-mcp-kit

Connect Claude (Claude Code / Claude Desktop / Codex — any MCP client) to **your own Telegram account** so it can read your chats and send replies. This kit is everything we wish we had on day one: it took us several hours and many broken iterations to get a stable setup; with this kit it should take you ~15 minutes.

Built and battle-tested at [Palo Alto AI Research Lab](https://github.com/tonydzi/tonydzi) — our Claude fleet reads and writes Telegram through exactly this setup, every day, across 5 machines.

What you get

A working setup path on top of the excellent upstream server `chigwell/telegram-mcp` (MTProto user-account, pure Python, no TDLib build pain).
`PROMPT.md` — a copy-paste prompt you give to Claude Code or Codex, and it performs the whole installation for you, including the gotchas.
Patches we run in production, pinned to upstream commit a008ac2 (apply cleanly there; run git checkout a008ac2 after cloning, or resolve conflicts yourself on a newer HEAD):
- `patches/0001-transport-sse-env.patch` — run the server as one shared HTTP/SSE daemon per machine instead of a stdio copy per session. With 10+ parallel agent sessions the stdio model wasted us ~2.7 GB RAM on duplicate Telegram clients; one daemon fixes that. Opt-in via MCP_TRANSPORT=sse (+ MCP_HOST/MCP_PORT), default behavior unchanged. Update: current upstream HEAD now ships env-selectable transport natively — on latest upstream skip this patch and just set the env vars; the daemon + watchdog in this kit work the same.
- `patches/0002-extra-tools-multiaccount.patch` — tools we found missing in daily agent work: search_dialogs (find private groups by title, not only @usernames), search_messages_global, get_new_messages_since (incremental polling for watchers), resolve_message_link