Agent Plugins 1.0: The open standard that unifies AI agent extension

What happened: a standard for packaging AI agent extensions is born

On February 25, 2025, Vercel announced the public availability of Agent Plugins 1.0.0, an open and neutral standard that defines how to package and distribute components that extend the capabilities of artificial intelligence agents. The specification, available at agent-plugins.org, was developed collaboratively by representatives from AWS, Anysphere, GitHub, Microsoft, OpenAI and Vercel, and has an initial technical committee made up of maintainers from these same companies.

The format focuses on two types of components: Agent Skills (reusable instructions) and MCP servers (Model Context Protocol, which connect to tools and services). Agent Plugins defines a standard directory structure: a plugin.json manifest at the root, a skills/ folder for skills, and an mcp.json file for MCP server configuration. This way, the same plugin can be discovered and loaded by any compatible client without the need for specific adaptations.

At launch, clients that already support Agent Plugins include ChatGPT, Codex, Cursor, GitHub Copilot, Kiro and VS Code, ensuring significant initial adoption.

Why it matters: the end of fragmentation in the agent ecosystem

Until now, developers of extensions for AI agents faced a fragmented landscape: each client (ChatGPT, Copilot, Cursor, etc.) had its own conventions for packaging Skills or configuring MCP servers, forcing the same component to be adapted to multiple formats. Agent Plugins solves this problem by offering a portable contract that specifies only what is necessary for a client to discover and load components, leaving installation, distribution and usage policies in the hands of each client.

This minimalist approach is key: the standard does not redefine Skills or MCP, but provides a common packaging layer. It also includes an extension mechanism with namespaces so that clients can add specific functionalities without contaminating the standard format. This allows innovation to continue on each platform while consolidating an interoperable core.

The participation of giants like OpenAI, Microsoft and AWS is no coincidence: they all have an interest in making AI agents easier to develop and deploy, and a common standard reduces integration costs and fosters a more vibrant plugin ecosystem.

Consequences: impact on developers, businesses and the future of work

For developers, Agent Plugins simplifies the creation and distribution of extensions: a single package works across multiple clients, reducing maintenance work and expanding potential reach. This could accelerate the proliferation of high-quality plugins, similar to what happened with browser plugins or VS Code extensions.

For businesses, standardization reduces vendor dependency and facilitates the adoption of AI agents in their workflows. By being able to reuse the same plugins across different tools, organizations can avoid lock-in and choose the best solution for each case. Furthermore, being an open standard encourages transparency and security, as plugins can be audited by the community.

In the future of work, interoperability is essential for AI agents to become ubiquitous assistants that act across multiple platforms. Agent Plugins is a step toward that vision, allowing an agent configured in one tool to move to another without friction.

What readers should know

First, Agent Plugins is an emerging standard: although it has the backing of major players, its success will depend on community adoption and the evolution of the technical committee. It is likely that we will see new versions that incorporate more component types, such as commands or hooks, as demonstrated needs arise.

Second, the specification is deliberately small and easy to implement, making it easy for other clients to join. If you are a developer of AI tools, consider adopting this format in your products. If you are a user, expect to see greater availability of plugins and a more consistent experience across platforms.

Finally, it is worth remembering that standardization also poses challenges: governance of the standard, moderation of malicious plugins, and possible premature standardization of certain components. The community must remain vigilant to ensure that the standard evolves in an inclusive and safe manner.

In summary, Agent Plugins 1.0 is a significant step toward the maturity of the AI agent ecosystem, with the potential to transform the way these technologies are developed and used.