Agents need a way to notify humans.
Not every task should stay hidden inside an IDE or terminal.
Sometimes an agent finishes a job, needs approval, hits a blocker or wants to send a generated artifact.
For that, I built MCP Telegram Agent.
Repo:
https://github.com/tecnomanu/mcp-telegram-agent
It is a small TypeScript MCP server over stdio.
The goal is simple: let any MCP-compatible AI agent send a Telegram message with one tool call.
What it includes:
- send Telegram notifications
- validate config before sending
- token-first setup
- guided onboarding flow
- setup code verification
- auto-generated MCP config snippet
- support for npx usage
The part I care about most is onboarding.
MCP tools often fail not because the tool logic is hard, but because setup is confusing.
A user should not have to guess the chat id, paste unsafe full API URLs or debug config by trial and error.
This project tries to make that path smaller.
Run with:
npx -y mcp-telegram-agent
If you build MCP servers, I think onboarding deserves as much design as the tool itself.
Top comments (0)