When you’re working with AI to build something, there’s a weird bottleneck that starts to emerge. The AI writes the code, but it has no idea what actually happened. It has a conceptual model of how things should work, but it can’t see the result.
For the first month or so of using AI heavily, I began to notice that tedium—copying console logs, server logs, DOM elements. You tell the AI something is wrong and try to give it additional context, then send it off to figure out what’s happening. The AI is working blind.
That’s the problem Relay Inspect solves.
What It Is
Relay Inspect is a lightweight MCP server that gives the AI tools to directly assess what’s happening in your browser.
It uses Chrome port forwarding to let the AI see the DOM, check applied styles, read the console and take screenshots. It also connects to your development server to view server logs and restart it if needed.
It works with Chrome or any Chromium-based browser.
In Practice
I’m a product designer. I like building. So I’ve naturally been drawn to these AI tools. At first I was so enamored that it was hard to see the flaws. Over time, the inefficiencies became more noticeable.
When tinkering on things—like my Hacker News extension—I started to feel the monotony. I was spending a lot of time copying and pasting logs and code for issues that would be pretty simple to solve if the AI wasn’t working blind.
Once I added Relay Inspect to my workflow, that dynamic changed. When it makes a visual change, it takes a screenshot, executes the update, then takes another screenshot and compares. If something breaks, it checks the server and restarts it. It’s gotten good at spotting the gap between what it expected and what actually happened.
Now most of the problems I run into aren’t visual bugs. They are places where I’ve been too vague in my instructions or my approach to something was just wrong.
Try It
Relay Inspect is open source. If you use Chrome or a Chromium browser, you can add it to your any Claude, Codex or OpenCode project as an MCP server.