Tech giants launch unified Agent Plugins standard

A coalition of major technology companies has launched Agent Plugins v1.0, a free and open standard designed to unify the fragmented ecosystem of AI agent workflows. Overseen by a Technical Steering Committee that includes Amazon, Cursor, Microsoft, OpenAI, and Vercel, the new specification establishes a single portable directory format. This standard allows developers to package Agent Skills and Model Context Protocol (MCP) server configurations together, ensuring they can run seamlessly across different AI clients without requiring custom rewrites for each platform.

The initial lineup of compatible launch clients includes ChatGPT, Codex, GitHub Copilot, VS Code, Cursor, and Kiro (AWS), all of which can now load the exact same plugin format. Under the hood, the specification simplifies integration by requiring only a basic directory structure. This includes a plugin.json file for identity and specification versioning, alongside a SKILL.md file containing instructions for the agent. Originally developed by Anthropic, these Agent Skills bundle instructions, scripts, and reference documents without needing a dedicated server. Additionally, the directory can house an mcp.json file to bundle MCP server configurations, which supports stdio, Streamable HTTP, and legacy SSE transports.

For AI practitioners and software engineers, this release marks the end of maintaining parallel versions of the same tool for different environments. Previously, an integration built for GitHub Copilot required significant modification before it could function in Cursor, while tools wired for Codex were incompatible with VS Code. By defining a minimal shared structure, Agent Plugins v1.0 keeps the portable elements of a tool consistent while allowing individual clients to retain full control over installation, permissions, and user experience. Developers can access the specification at agent-plugins.org and participate in public governance and proposals on GitHub.