Vercel, OpenAI and Microsoft Launch Agent Plugins 1.0 to End AI Vendor Lock-in
A new 'write-once-run-anywhere' specification allows AI agent skills to be ported across major platforms including ChatGPT and VS Code.
Vercel, in collaboration with Amazon, Cursor, Microsoft, and OpenAI, has released the Agent Plugins 1.0 specification. The new standard establishes a universal container for AI agent skills and tools, enabling developers to deploy capabilities across different agent platforms without rewriting code.
The specification implements a "write-once-run-anywhere" architecture by bundling AI skills and tools into a standardized directory structure. Each plugin consists of a plugin.json manifest, a dedicated skills sub-directory, and an mcp.json configuration file. To ensure broad connectivity, the spec integrates the Model Context Protocol (MCP) and the Agent Skills standard originally developed by Anthropic in 2025.
Several major AI clients already support the standard, including ChatGPT, GitHub Copilot, VS Code, Cursor, Codex, and Kiro. The project is managed through an open governance model overseen by a technical oversight committee and is licensed under Creative Commons Attribution 4.0 International (CC-BY-4.0). While the initiative is supported by the Linux Foundation’s Agentic AI Foundation (AAIF), it operates as an independent entity. In a move toward further industry consolidation, Google has folded its Agent2Agent (A2A) communications protocol effort into the AAIF.
The Push for Modularity
Until now, the agentic AI landscape has remained heavily fragmented. Developers were forced to build platform-specific tools and hooks for every individual AI environment they targeted, creating significant friction and vendor lock-in. This release is part of a broader industry shift toward modularity, complementing other efforts such as Google's Agentic Resource Discovery and the AAIF's "Skills Over MCP" project, which aims to bundle skills directly into the MCP server.
Implications for the Ecosystem
By standardizing how agent capabilities are packaged, the industry effectively reduces the cost of switching between AI providers. A specific "superpower"—such as a specialized data analysis tool or a niche API integration—created for one platform can now be instantly deployed on another. This interoperability is expected to accelerate the growth of the agentic ecosystem by allowing developers to reach a wider audience of users across multiple LLM interfaces simultaneously.
Future Outlook
As the specification gains adoption, the focus will shift toward how these portable plugins interact in multi-agent environments. The Agentic AI Foundation noted that the goal is to keep the shared format predictable without requiring it to "absorb every client-specific experiment happening across agent products." Observers will be watching to see if other major cloud providers adopt the standard to prevent a new divide between supported and unsupported agent ecosystems.