Homework 3

Talk to your agent from your phone, about your research

Due before class on Wednesday, October 21, 2026

Your week 2 harness gets its first two plugins: a Telegram MCP server, and an MCP server for your own research. When you're done, you ask your agent a research question from your phone, it works on your computer, and it answers in the chat. Your harness keeps only the loop and bash; everything else can arrive as a plugin.

You need your week 2 harness and its model API, Telegram on your phone, and uv. Your AI coding agent writes most of the code with you.

Missed week 2? That's fine. Build both plugins and use them from an agent you already have (Claude Code, Codex or OpenClaw) instead of your own harness. But building on your own agent loop teaches you considerably more, so if you can, finish week 2's harness first.

Start

  1. Open an AI coding agentClaude Code, Codex, Cursor, GitHub Copilot, … in your week 2 harness folder.
  2. Give it this Read https://class.xhelio.ai/week-3/homework.md and help me with my homework.
  3. Build it togetherIt helps you create the bot, builds both servers and plugins with you, extends your harness, tests it from your phone, and opens your pull request.

What you build

You may use the official MCP Python SDK. No agent frameworks.

Safety

A bot connected to bash on your computer has all three legs of the lethal trifecta: your files, messages from the internet, and a way out.

Messages to a Telegram bot are not end-to-end encrypted. Don't send anything secret through it.

What you submit

A pull request adding both plugin folders under week-3/submissions/<your-github-username>/: <you>-telegram/ and <you>-<name>/, plus a README.md that links to your harness repository and says how to run it in Telegram mode.

It counts when your harness repository is public and updated, both plugins work in your harness and in Claude Code or Codex, you can chat, approve commands and get a research answer from your phone, messages from anyone else are ignored, and no token or key is committed. Say in your pull request which AI helped you, and how.