I've already seen tools that generate MCP servers. So... Why can't the LLM just do it by itself then?

I'm genuinely curious, as right now it just feels like we are putting a non-deterministic layer on top of strict API specifications that are not AI-only, so anyone can integrate with them.

Wasn't the point of AI to overtake human work? Now we are just writing code for it? 🤔

Are there any public benchmarks/studies that prove/explain how and why MCP servers are better than giving an LLM your OpenAPI, etc. specs?

OAuth = email+password, but delegated to someone who cares about it more than you could.

(They are trying to discover an open redirect vulnerability in your app)

Hi, this means someone is actually trying to hack your users, but Auth.js is mitigating them. Maybe time for setting up a firewall/bot protection!

💚

If you use auth with refresh tokens that have automatic reuse detection. I.e. if an auth token is expired, you can request a new one using the refresh token only once. How do YOU on your site deal with the race condition where opening multiple tabs/requests at once logs the user out? Not at all?