ClaudeBridge

Claude Code CLI only really knows one shape: one terminal, one login, one human typing. The moment I wanted to call it from a script instead - a cron job, a small automation - there was nothing clean to hand it, since that OAuth login doesn't come with an API key attached. Since I'm lazy like that (lol 😂), I built ClaudeBridge: log in once through a browser (real Claude OAuth, not an API key), generate a key from the same page, and POST /ask just works from anywhere after that. Each login lives in its own isolated folder too, so nothing bleeds between accounts if more than one person uses it.

Note: this is built around a subscription I already pay for, purely for my own automation — not a way around Anthropic's official API pricing. For production or multi-user stuff, go through the real Claude API.

Tech Stack:

• Express & Node.js (API routing, spawning the claude CLI)
• MySQL (metadata only — never a raw token)
• Vanilla HTML / CSS / JS (no framework, no build step)

Interesting bits:

1. An API key doesn't hold a credential — it's just a lookup that points to the right session folder for the CLI to use.
2. /ask runs fully locked down — no tool or file access, plain text in, plain text out.
3. There's no claude models command, but the same OAuth token also works against Anthropic's public /v1/models — so the docs page shows your real model list live instead of a hardcoded guess.