Day 0

I wasn't trying to start a company.

I was just stuck.

I'd been using Claude Code for a few weeks. Loved it. It was changing how I worked. But that afternoon, something wasn't working and I couldn't figure out why.

Claude kept asking for logs. And I kept copy-pasting them. Badly. Partial outputs. Missing context. Wrong timestamps.

We were going in circles.

The frustration

Here's the thing about debugging with AI—it's only as good as the information you give it. Claude can reason through almost anything. But it can't see what it can't see.

And I was the bottleneck.

I'd run a command, scroll through output, copy what I thought was relevant, paste it in. Claude would ask a follow-up. I'd go back, copy more. Miss something. Go back again.

It was slow. It was frustrating. And it hit me: this is stupid.

Claude should just... see the logs.

The lightbulb

I knew about MCP—Model Context Protocol. Anthropic had just released it. The idea is simple: give Claude tools it can call directly. Instead of you being the middleman, Claude can reach out and grab what it needs.

So I built a thing. Nothing fancy. Just a small MCP server that watches log files and lets Claude query them directly.

I called it "the log watcher."

Took me a few hours. And suddenly, debugging was different.

Claude: "Let me check the logs from the last 5 minutes."

Actually checks the logs.

Claude: "I see the error. It's on line 47. Here's what's happening..."

No more copy-pasting. No more "can you show me the full output?" No more being the bottleneck.

I didn't want to sell this

My first instinct wasn't "startup idea." It was "I should share this."

Because if I had this problem, other developers did too. And the solution was simple enough that it felt wrong to keep it to myself.

Open source it. Put it on GitHub. Move on.

But then I started thinking.

What else is Claude missing?

Logs were just one thing. What about errors? What about metrics? What about the database state, the request history, the deployment status?

Every time Claude asked me for context, it was a sign. A gap. Something it should be able to see but couldn't.

I started making a list:

• Logs ✓ (just built this)
• Errors and exceptions
• Application metrics
• Database queries
• HTTP requests
• Background job status
• Feature flags
• Cache state

The list kept growing.

And I realized: this isn't one tool. This is a whole layer of infrastructure. The eyes and ears that AI needs to actually help with real work.

Not a company yet

I want to be clear—at this point, I wasn't thinking about an AI-run company. That came later.

I was just a developer who wanted Claude to be more useful. Who was tired of being the bottleneck. Who saw a gap and wanted to fill it.

The log watcher was day 0. The spark.

It was ugly. But it worked.

Everything else grew from there.

— Andres