OpenAI and Rivals Publish Agent Plugins: One Package Format for AI Agents

OpenAI, Amazon, Microsoft, Cursor, and Vercel just published Agent Plugins 1.0, an open standard for packaging AI agent extensions. The pitch: build once, run anywhere. Instead of each tool expecting different folder layouts and setup, Agent Plugins defines a single package format.

A plugin is just a folder with a plugin.json file at its root. It bundles two things developers already use:

• Model Context Protocol (MCP) servers — connect agents to live tools and data.
• Agent Skills — reusable sets of instructions.

At launch, ChatGPT, Codex, Cursor, GitHub Copilot, Kiro, and VS Code support the format. The spec is openly licensed; no single company's roadmap steers it.

Deliberately minimal

The standard intentionally avoids the hard parts. Marketplaces, installation, permissions, sandboxing, and trust remain client-specific. That keeps adoption low-friction, but it also means deciding whether a plugin is safe to run is still every client's job — a live concern after fake Agent Skills slipped past security scanners earlier this year.

Not everyone is sold. Dax Raad of SST called it “a thin standard” whose useful parts will end up in client-specific extensions anyway. Others like Angie Jones welcomed it: “We neeeeded this.”

What it means for you

If you build agent skills or MCP servers, you can now package them once and reach a wide range of tools. The competitive landscape shifts too: a shared format helps small devs reach major clients, but it also cements whichever clients already have the users. The plumbing is agreed — the fight over marketplaces and trust is just beginning.