Introducing Agent Plugins

Categories: AI, Product

Summary

Major tech companies (AWS, Microsoft, OpenAI, Vercel, GitHub, Cursor) unified behind Agent Plugins—a vendor-neutral standard for packaging AI agent extensions. One manifest format replaces fragmented plugin ecosystems, reducing platform-specific work for developers and enabling seamless capability sharing across competing AI products.

Key Takeaways

  1. Agent Plugins standardizes on a single plugin.json manifest at folder root, eliminating multiple competing formats and reducing friction for plugin authors building across different platforms.
  2. The spec focuses on two immediate use cases: agent skills (reusable instructions/workflows) and MCP servers (agent-to-tools/data connectors), covering the most common extension patterns.
  3. Industry coordination across 6 major vendors (AWS, Cursor, GitHub, Microsoft, OpenAI, Vercel) signals a shift toward interoperable AI agent ecosystems rather than proprietary lock-in.
  4. The standard intentionally stays vendor-neutral by standardizing only packaging and discovery—leaving marketplaces, permissions, and runtimes to individual products, enabling adoption flexibility.
  5. Plugin developers get immediate portability: one version format works across multiple agent products (skills, MCPs, or both), dramatically reducing maintenance overhead.

Related topics

Transcript Excerpt

Skills and MCP servers are making agents more capable, and plugins give developers a way to package those capabilities so they can be shared and reused. But today, every agent product expects a different manifest, folder structure, and setup process. That's why contributors from AWS, Cursor GitHub Microsoft OpenAI and Vercel came together to create Agent Plugins, an open, vendor-neutral way to package extensions for agents. One format for people building plugins, and one predictable way for agent tools to find them. At its simplest, an agent plugin is just a folder, with a manifest called plugin.json at the root. This first release focuses on two things developers already use, agent skills for reusable instructions and workflows, and MCP servers, which connect agents to tools and data. You…

More from OpenAI

Featured in