The Universal Standard: How Every AI Company Adopted Agent Skills

In the span of eighteen months, the AI industry achieved something remarkable: near-universal adoption of a common standard for agent skills. This is the story of how it happened, why it matters, and what it means for the future of AI development.

The Fragmentation Problem

In early 2024, every AI company had its own approach to extending agent capabilities. OpenAI had function calling and the Assistants API. Anthropic had tool use with XML-based schemas. Google had Vertex AI extensions. Each system was powerful but incompatible. A tool built for one platform could not work on another without significant rewriting.

This fragmentation was costly. Developers had to maintain multiple versions of the same capability. Enterprises hesitated to invest in tooling that might lock them into a single provider. And the broader ecosystem suffered from divided attention and duplicated effort.

The MCP Breakthrough

The turning point came when Anthropic open-sourced the Model Context Protocol (MCP) in late 2024. MCP was not just another proprietary format — it was designed from the ground up as an open standard. It defined how skills declare their capabilities, how agents discover and invoke them, and how results are returned.

What made MCP different was its pragmatism. Rather than trying to define every possible interaction, it focused on the 90% case: a skill has tools (functions the agent can call), resources (data the skill can provide), and prompts (instructions for the agent). This simplicity made adoption straightforward.

The Domino Effect

Once MCP gained traction, the dominoes fell quickly. OpenAI announced MCP compatibility in their Assistants API in early 2025. Google followed with Gemini MCP support. Microsoft integrated it into Copilot Studio. Within months, every major AI platform either natively supported MCP or provided compatibility layers.

The enterprise adoption was equally swift. Companies like Stripe, Cloudflare, and HashiCorp began publishing official skills for their platforms. Vercel Labs released React and web development skills. Sentry created monitoring skills. The ecosystem grew from dozens to hundreds to thousands of skills in under a year.

What Standardization Enabled

The universal standard unlocked several capabilities that were impossible in the fragmented world:

Portability. A skill built by Anthropic works with OpenAI's models and Google's Gemini. Developers write once and deploy everywhere. This dramatically reduced the cost of skill development and increased the addressable market for skill creators.

Composability. Skills from different creators can work together seamlessly. An agent can use a Stripe payment skill alongside a Sanity CMS skill and a Cloudflare deployment skill — all in the same workflow. The standard interface makes this composition natural and reliable.

Security. A common standard enabled common security tooling. Marketplaces like CastKit can scan skills against a known schema, verify their behavior, and flag anomalies. The Trail of Bits security skills, for example, can audit any MCP-compatible skill regardless of its creator.

Discovery. With a standard format, skills became searchable and comparable. Marketplaces could categorize skills by capability, rate them by quality, and recommend them based on use case. The rarity system — Common, Rare, Epic, Legendary — emerged as a community-driven quality signal.

The Current Landscape

As of early 2026, the numbers tell the story. Over 116 verified skills from major tech companies are available on CastKit alone. Community contributions add hundreds more. The total install count across all skills exceeds two million. And the security scanning infrastructure has caught over 341 malicious skills before they could cause harm.

The companies that moved fastest — Anthropic with its 16 official skills, OpenAI with nearly 30, and the community with its diverse contributions — have established themselves as the skill ecosystem leaders. Their skills serve as reference implementations that others learn from and build upon.

Looking Forward

The standardization of agent skills is not the end of the story — it is the beginning. The next challenges include real-time skill composition, cross-agent skill sharing, and skill versioning at scale. But the foundation is solid, the ecosystem is thriving, and the direction is clear.

The universal standard did not just make AI agents more capable. It made the entire AI ecosystem more collaborative, more secure, and more accessible. And that might be the most important development in AI since the transformer architecture itself.