The MCP Bet

In November 2024, Anthropic quietly released something called MCP.

Most people ignored it. "Another protocol," they shrugged. "Another standard that'll go nowhere."

I read the spec and couldn't sleep that night.

What I saw

MCP is deceptively simple: a standard way for AI to talk to tools.

But think about what that enables.

Before MCP, if you wanted Claude to access your database, you had to build a custom integration. Handle authentication yourself. Parse responses yourself. Hope it all worked.

With MCP, you expose a standard interface. Claude connects. It just works.

I realized: every product I'm building could have an MCP interface. And suddenly, Claude could access everything—logs, errors, metrics, secrets—through a unified protocol.

That's not a feature. That's a paradigm shift.

The timing was perfect

I'd been thinking about AI-powered observability for months. The idea was good, but the implementation was messy. Too many integrations. Too much glue code.

MCP solved all of it.

Instead of building custom integrations for each AI model, I build one MCP interface per product. Any AI that speaks MCP can connect.

The architecture went from complicated to elegant overnight.

Why I'm going all-in

Here's my bet: MCP becomes the standard way AI talks to tools.

Not because it's technically superior (though it is). Because it's backed by Anthropic, and Claude is increasingly the AI developers want to use.

If Claude uses MCP, developers will build MCP tools. If developers build MCP tools, more AI products will support MCP. Network effects take over.

In two years, "does it have MCP support?" will be like "does it have an API?" today. Table stakes.

The competition can't catch up

Datadog, Sentry, New Relic—they're all SaaS-first. Their architectures assume your data lives in their cloud.

MCP works best with local connections. Fast. No rate limits. No API keys to manage.

Their architecture is a liability. Mine is an advantage.

By the time they figure out how to retrofit MCP onto their cloud-first products, I'll have a two-year head start building native MCP tools.

What if I'm wrong?

MCP could fail. Another standard could win. Anthropic could abandon it.

But I don't think so.

MCP is simple. It solves a real problem. It's already shipping in Claude Desktop. Real developers are building real tools.

And even if MCP doesn't become the universal standard, the products still work. MCP is just one interface. There are also REST APIs, SDKs, webhooks.

I'm betting on MCP, not depending on it.

The moment I knew

The first time I asked Claude "what errors happened today?" and it actually pulled data from my Reflex instance via MCP—no copying, no pasting, just a direct answer—I got chills.

That's the future. AI that can actually do things. Not just chat about things.

MCP makes that possible. And I'm building everything around it.

— Andres