BriefLookout

Technology & AI · Demonstration

The quiet standardization of model context protocols

Three competing standards for letting AI agents use tools converged in under a year. What that says about who actually controls the agent ecosystem.

BriefLookout · 8 Aug 2026 · 6 min read

A year ago, three separate specifications existed for how an AI model requests and uses external tools — a calendar, a database, a piece of internal software. Each was backed by a different major lab, and developers building agent software had to pick one and accept the lock-in.

That's mostly settled now. The specification with the broadest early adoption absorbed most of the features unique to its rivals, and the labs that backed the other two have shipped compatibility layers rather than continuing to maintain competing standards.

What made the difference wasn't technical superiority so much as which standard third-party tool makers built support for first. Once a critical mass of tools (project trackers, internal databases, scheduling software) supported one protocol, switching costs for everyone else rose fast.

This is a familiar pattern from earlier standards fights, but the speed was unusual: it typically takes several years for a developer ecosystem to consolidate around one specification. Here it took under twelve months.

Why it matters

Whoever's protocol becomes the default effectively sets the rules for how every AI agent connects to outside software — a structurally important, low-visibility layer of the emerging agent ecosystem.

What to watch

  • Whether the losing standards' compatibility layers get maintained long-term or quietly deprecated
  • Enterprise software vendors' choice of which protocol to build native support for first
  • Whether a fourth, security-focused variant emerges in response to early exploits reported against the dominant standard
Sources (2)

Related reading

← Back to Technology & AI