Agent Plugins - one package format for skills and MCP servers

Canonical version: Agent Plugins - one package format for skills and MCP servers.

A new open specification landed: Agent Plugins 1.0, with an initial technical steering committee made of Amazon, Cursor, Microsoft, OpenAI and Vercel. It defines a portable package format for extending agents, so an extension is authored once instead of being repackaged per client.

The format is deliberately small. A plugin is a directory with a plugin.json manifest at its root. Two component types are standardized: AI Agent Skills discovered from a skills/ folder (each with its own SKILL.md, following the Agent Skills spec), and Model Context Protocol (MCP) servers declared in mcp.json, over stdio, Streamable HTTP, or the legacy HTTP+SSE transport. Anything client-specific goes under a reverse-domain extensions namespace.

The spec does not cover distribution, installation, permissions, or UX; those stay with each client.

GitHub Copilot shipped support in VS Code, the Copilot CLI and the Copilot app; ChatGPT, OpenAI Codex, Hermes Agent, OpenClaw, Cursor and Grok are on the list too.

Agent Plugins overlaps with Microsoft's own APM (AI) and a handful of other manifests, and none of them has won yet. I think the direction is right: the value of a skill or an MCP server should not depend on which agent you happen to run.

References

• Specification: https://agent-plugins.org/specification
• Site: https://agent-plugins.org/
• Repository: https://github.com/agentplugins/agent-plugins-spec
• GitHub Copilot changelog: https://github.blog/changelog/2026-08-12-agent-plugins-1-0-in-vs-code-copilot-cli-and-the-copilot-app/
• Discussion on overlap with other formats: https://github.com/agentplugins/agent-plugins-spec/discussions/36

Related

• AI Agent Skills
• Model Context Protocol (MCP)
• APM (AI)
• Agent Plugins Specification

About Sébastien

Ready to get to the next level?