It reads like a simple book, not documentation. We start with the chaos—the M×N integration nightmare that has engineers debugging at 3AM. Then walk through how MCP solves it for agents, piece by piece.
It's a tribute to the protocol and a reference for anyone building MCP ecosystems.
If you're designing AI-to-tool connections, we hope it saves you some of the headaches we had.
Not sure if this is worth sharing, but I've been using AI coding assistants heavily for the past few months and kept running into the same frustrating pattern.
I'd have these amazing flow sessions with Claude or other AI tools where we'd build something that felt brilliant. The code looked clean, the architecture seemed solid, and I'd go to bed feeling productive.
Then I'd wake up and actually try to use what we built. Half the functions were just sophisticated-looking stubs. Error handling that caught exceptions just to ignore them. TODOs that were more like "TODO: figure out how this should actually work."
The worst part wasn't that the AI was wrong - it was that the AI was convincingly wrong. In the moment, everything felt right because the code looked professional and the comments were confident.
So I started building this tool called "sniff" (yeah, like sniffing out BS) to catch these patterns in real-time. It looks for things like:
* Functions that claim to do X but actually just return a default value
* Error handling that's all ceremony and no substance
* Comments that overpromise what the code delivers
The weird part was using AI to help build the tool that catches AI mistakes. Meta level stuff where sniff would analyze its own improvements and flag them as suspicious. "Your new feature detection is just an untested regex" -thanks, tool I just wrote.
I've been using it for months now and it's honestly changed how I work with AI assistants. Still get the creative benefits but with a reality check built in.
Anyway, I open sourced it in case anyone else has dealt with this. Maybe it's just me overthinking things, but figured I'd share: https://github.com/conikeec/sniff
Not trying to solve world hunger here, just scratching my own itch. Let me know if you've had similar experiences with AI coding tools - curious if this resonates with others or if I'm just paranoid about my own code.
I have the second pair you listed, and find them terribly uncomfortable to wear for more than a half an hour or so. I've tried stretching them on a medium-sized cardboard box but didn't notice much improvement. They make my head feel like it is being crushed, and give me a terrible headache rather quickly. They crush my ears into my skull too. The noise canceling and battery life on them are great, but if you considering purchasing these and have a slightly larger-than-average sized head, this is something you should consider.
I'm getting a 400 error when I try to complete the 'choose a username' screen.
No visible reaction to pushing the continue button at all, but the error shows up in chrome's console.
"Uncaught SyntaxError: Unexpected token O"
Chrome 25 on windows 7.
I just tried again from scratch, without going through the google oAuth page, and its worked this way. No idea what went wrong the first time around.
Hardest and most rewarding was all the serious engineering we did to make the app and gestures fun and fast. Most of the components in the app are custom, from the navigation controller to the story menu down to the scroll physics. Aria just answered this Quora question that goes into some of the details.
Not just building on it—understanding it. The architecture decisions. The tradeoffs. The "why" behind the protocol.
Today we're releasing that learning : https://makingmcp.com/
It reads like a simple book, not documentation. We start with the chaos—the M×N integration nightmare that has engineers debugging at 3AM. Then walk through how MCP solves it for agents, piece by piece.
It's a tribute to the protocol and a reference for anyone building MCP ecosystems. If you're designing AI-to-tool connections, we hope it saves you some of the headaches we had.
Grab a coffee and enjoy!