Why AI Agent Infrastructure just got safer to invest in

MCP Just Had Its USB-C Moment

If you've been waiting for a sign that AI agent infrastructure is ready for real investment — this is probably it.

What happened

On December 9th, Anthropic donated MCP — the Model Context Protocol — to the Linux Foundation. They announced the formation of the Agentic AI Foundation, with OpenAI, Google, Microsoft, AWS, Snowflake, and about a dozen other companies as founding members.

OpenAI donated AGENTS.md. Block donated Goose. Competitors agreeing on infrastructure standards. That almost never happens.

I've been building with MCP since summer. And honestly, this news changes the math on whether it's worth investing in.

What MCP actually is

MCP is how AI models connect to your tools and data. Think of it as a universal adapter — one protocol that lets AI agents talk to Slack, GitHub, Postgres, Salesforce, Google Drive, whatever.

Before MCP, every integration was custom. Build something for Claude, rebuild it for GPT, rebuild it again for Gemini. Or pick one vendor and hope you made the right bet.

That's where the USB-C analogy comes in.

Remember when every phone had a different charger? You'd buy cables and accessories knowing that your next device might make them useless. USB-C standardization fixed that. One connector, many devices, no lock-in.

MCP under the Linux Foundation does the same thing for AI integrations.

Why open governance matters

Until last week, MCP was Anthropic's project. Good technology, but still controlled by a single vendor.

Now it's under open governance. A Technical Steering Committee with representatives from multiple companies. No single vendor controls the roadmap.

This matters if you're making infrastructure decisions that need to last more than 18 months. Which, for most mid-market companies, is basically all infrastructure decisions.

The ecosystem is already mature. 97 million SDK downloads per month. Already adopted by ChatGPT, Gemini, Copilot, Cursor, VS Code, JetBrains. This isn't a bet on something that might become a standard. It's recognition of something that already is.

What this means practically

For mid-market companies exploring AI agents, a few things shift.

Your integrations aren't vendor-locked anymore. Build an MCP server once, it works across Claude, GPT, Gemini, and Copilot. If you switch AI providers next year, your integrations come with you.

You don't have to build everything from scratch. Pre-built MCP servers already exist for Slack, GitHub, Postgres, Google Drive, Salesforce, and a growing list of others. The plumbing is increasingly commodity.

The "wrong AI bet" risk got smaller. This has been the quiet anxiety behind a lot of AI project delays. What if we invest in Claude and GPT-5 changes everything? What if we build for OpenAI and Anthropic pulls ahead? With a shared protocol layer, the risk of that bet going sideways drops significantly.

Where we are with it

We've been running MCP in production for about 6 months now. Building custom servers for client-specific tools. Connecting agents to internal systems. Figuring out what works and what doesn't.

A few things we've learned:

The pre-built servers are good starting points, but most real implementations need customization. The protocol itself is solid — it's the "what should this agent actually be able to do" questions that take time.

Security and permissions matter more than you'd think. Just because an agent can connect to a system doesn't mean it should have access to everything in that system. We spend a lot of time on scoping.

Start smaller than you think. The temptation is to build an agent that can do everything. The reality is that agents that do one thing well are more useful than agents that do ten things poorly.

Where this is heading

I think we're at the point where "should we explore AI agents" is becoming "how should we implement AI agents." The infrastructure layer is maturing. The standards are emerging. The risk profile is changing.

MCP under the Linux Foundation is a signal that the industry is moving from experimentation to infrastructure. That's usually when mid-market companies should start paying closer attention.

If you're exploring agents and want to talk through what we've learned — the stuff that worked, the stuff that didn't — happy to share. Reach out.