// the take
A weekend-scale experiment from Eugene Yan showing how to orchestrate multiple AI agents from the terminal: Amazon Q CLI drives the agents, MCP exposes RSS parsing as tools, and tmux splits the screen so you can watch each sub-agent fetch and summarize a chunk of news feeds in real time. It's for developers curious about multi-agent patterns and MCP tool design, not for anyone who wants a maintained news reader.
The main-agent/sub-agent split (grab feeds, chunk them, spawn workers, collect results) is a clean, legible example of fan-out/fan-in agent orchestration that's easy to adapt to other batch-processing tasks. Using tmux panes to show each agent's live output is a genuinely useful debugging trick — you see what each agent is doing instead of staring at one interleaved log. The per-source fetchers (hackernews.py, techcrunch.py, wsj.py, etc.) are small and easy to read end to end in a few minutes. MCP is used as intended here — RSS fetch/parse exposed as tools the agent calls, not baked into a giant prompt.
It's tied to Amazon Q CLI as the agent harness, so you need an AWS account and Q access just to run it — a heavier dependency than a plain API key, and it won't port cleanly to other agent runtimes. There's no visible error handling in the fetchers; one dead or rate-limited feed likely stalls or breaks its whole chunk. The orchestration is hardcoded to 3 equal-sized chunks regardless of feed count or size, so it doesn't rebalance if one source is much slower than the others. Last pushed July 2025 with no tests or CI, and the author calls it a 'little experiment' — this is a demo to learn from, not something built to keep running unattended.
We read the README, the directory tree and a key source file in October 2026. Nobody pays to be listed. How we judge →