How is Agent Plugins different from MCP?

MCP (Model Context Protocol) is the connection layer that lets an agent reach live tools and data. Agent Plugins is the packaging layer around it — it standardizes where the MCP config and Agent Skills live so a whole extension is portable. Agent Plugins wraps MCP; it does not replace it.

Quick Answer

What Is the OpenAI Agent Plugins Standard? (2026)

Q: What is the OpenAI Agent Plugins standard?

Agent Plugins is an open, vendor-neutral packaging standard for AI-agent extensions, published August 6, 2026. A single portable package (with a required plugin.json manifest) bundles Agent Skills and MCP server configs so one extension runs across ChatGPT, Cursor, Copilot, Codex, VS Code, and Kiro — 'build once, run anywhere.'

Q: Who backs Agent Plugins?

The 1.0.0 spec was published by a steering committee including OpenAI, Amazon, Cursor, Microsoft, and Vercel, with Google joining as a core maintainer. Launch-compatible clients include VS Code, Cursor, GitHub Copilot, OpenAI Codex, and Kiro.

Published: August 8, 2026

The Short Answer

Agent Plugins is an open, vendor-neutral packaging standard for AI-agent extensions, published August 6, 2026. It defines one portable package format — a required plugin.json manifest plus fixed locations for Agent Skills and MCP server configs — so a single extension runs across compatible AI products. The pitch is “build once, run anywhere.”

Who Is Behind It

The Agent Plugins 1.0.0 spec was published by a steering committee including OpenAI, Amazon, Cursor, Microsoft, and Vercel, with Google joining as a core maintainer. Launch-compatible clients include VS Code, Cursor, GitHub Copilot, OpenAI Codex, and Kiro (support varies by component and MCP transport).

What a Plugin Bundles

A plugin package pairs two things that already had open specs and wide adoption:

Component	What it is
Agent Skills	Reusable packages of expertise — instructions, workflows, and context that teach an agent to do a specific task.
MCP servers	Model Context Protocol connections that let an agent pull live data and act on external tools.

The standard fixes where the MCP config lives and how a client maps it into its runtime, so the same folder works everywhere instead of needing a per-tool layout.

The Problem It Solves

Before this, every agent client wanted a different setup and folder layout. Moving a capability from ChatGPT to Claude to Cursor meant re-packaging it each time. Agent Plugins standardizes the package and discovery layer so that friction goes away.

What It Deliberately Does Not Cover

Agent Plugins is a low-level packaging standard, not a marketplace. It does not define installation, security, permissions, distribution, or UX rules — those are left to each client. It’s distinct from OpenAI’s Plugin Directory, which handles distribution of workflow packages.

Why It Matters

Standardizing packaging is how ecosystems compound: one skill or MCP integration now ships to five+ clients instead of one. Combined with MCP (the connection layer) and Agent Skills (the expertise layer), Agent Plugins is the missing distribution wrapper for portable agent capabilities.

Sources

9to5Mac — GPT-5 turns one; OpenAI shares new Agent Plugins standard (Aug 6, 2026): 9to5mac.com
The Next Web — OpenAI’s Agent Plugins open standard (Skills + MCP): thenextweb.com
Cybernews — Agent Plugin standard (OpenAI, Google, Amazon): cybernews.com