I built PocketLantern to help AI coding agents catch technical blockers

I've been using AI coding agents a lot, and kept running into the same problem: they often give plausible answers on technical choices, but miss blockers like EOL dates, breaking changes, vendor lock-in, pricing shifts, and migration constraints.

So I built PocketLantern.

PocketLantern is a local MCP server with bundled decision cards grounded in source-linked, time-sensitive facts. It’s not generic docs search. The goal is to help the agent catch blockers it would otherwise miss.

A few examples of the kinds of questions it helps with:

• Which auth provider should I choose
• Should I upgrade to Next.js 16
• Can I keep using this API path, or is it being removed

You can try it locally:

npm install -g pocketlantern
pocketlantern init

Then connect it to Claude Code and start asking technical decision questions.

GitHub: https://github.com/pocketlantern/pocketlantern

I’d especially love feedback on two things:

1. Is the framing immediately clear?
2. Does the first-run experience feel simple enough?