Five major tech companies—Amazon, Cursor, Microsoft, OpenAI, and Vercel—have agreed on a unified standard for AI agent plugins called Agent Plugins, which uses a single package format (plugin.json) to let developers write extensions once and reuse them across different platforms.
Previously, each product required its own formats and setup, making it costly for developers to support multiple platforms.
The standard covers packaging and discoverability and is being developed publicly on GitHub, though it does not yet include marketplace or permissions specifications.
What happened
Amazon, Cursor, Microsoft, OpenAI, and Vercel have created Agent Plugins, an open standard that defines a single package format for AI agent extensions. The standard uses a directory with a manifest file called plugin.json and supports two components: Agent Skills (reusable instructions and workflows) and MCP servers (connections to tools and data).
Why it matters
Until now, each product used its own formats and setup processes, forcing developers to rebuild extensions for every platform. A shared standard lets developers bundle and reuse their work across multiple platforms, removing friction for extension creators.
What to watch
The spec is being developed openly on GitHub and currently covers only packaging and discoverability—it does not yet address marketplaces, permissions, or runtime environments. Notably, Anthropic, which created the Model Context Protocol and Agent Skills standards, is not part of the effort despite having added its own plugin system to its Cowork tool.
Amazon, Cursor, Microsoft, OpenAI, and Vercel announced Agent Plugins, an open standard designed to unify how AI agent extensions are packaged and shared. The standard addresses a long-standing pain point: until now, each platform required developers to use its own folder structure, configuration format, and setup process. This forced developers to rebuild the same extension multiple times if they wanted to support multiple agents, creating barriers to adoption and slowing extension ecosystem growth.
Agent Plugins standardizes this process around a simple directory structure anchored by a manifest file called plugin.json. Version 1.0.0 defines two core components. Agent Skills are reusable instructions and workflows—the logic of what an extension does. MCP servers are bridges that connect agents to external tools and data sources. This two-part design separates the instruction layer from the connectivity layer, allowing flexibility in how extensions are composed.
The standard's scope is deliberately focused on the foundational layer: packaging and discoverability. It does not attempt to standardize marketplaces (where extensions are found and distributed), permissions (who can do what with an extension), or runtime environments (how different agents actually execute extensions). This narrow scope makes the standard easier to align on and less likely to hit stumbling blocks, while still allowing each platform to differentiate on marketplace features, security policies, and execution models.
The specification is being developed openly on GitHub, reflecting the partners' commitment to community input and transparency. However, one notable absence is Anthropic, the company behind both the Model Context Protocol and the Agent Skills standard that Agent Plugins incorporates. Anthropic recently added its own plugin system to Cowork, its desktop tool for agentic knowledge work, but is not part of the Agent Plugins coalition. The reasons for Anthropic's non-participation are not stated in available materials.
The Agent Plugins initiative addresses a real friction point in the emerging AI agent ecosystem: the fragmentation of extension formats across competing platforms. Each vendor—Amazon, Microsoft, OpenAI, Vercel, and Cursor—has built or is building agent tooling, yet they had no way for developers to write once and deploy everywhere. By agreeing on a shared manifest structure and package format, these five companies are making a bet that reducing developer burden will accelerate ecosystem growth.
The standard is deliberately narrow: it specifies only how to package and discover plugins, deferring harder questions like permissions, marketplaces, and runtime behavior to a later phase or to individual platforms. This pragmatic scoping increases the chances of adoption while leaving room for platforms to differentiate on commercial and security features.
One striking absence is Anthropic, which created both the Model Context Protocol (MCP) and Agent Skills as open standards and recently launched its own agentic tool, Cowork. Despite Anthropic's foundational role in these standards, it did not join the coalition. This gap may reflect either independent strategy or unresolved technical or commercial alignment questions.
AI-summarized, only the topics you pick — one digest a day via Email, Slack, or Discord.
Free · takes 30 seconds · unsubscribe anytime
Ask AI anything about this article. Q&As are published on this page for other readers too.
Security researchers led by Alexander Panfilov discovered a vulnerability in the APIs of all major AI provider…
CEO Sundar Pichai announced that the Gemini app has surpassed 1 billion monthly active users, making it the 14…
River AI, founded by xAI co-founder Igor Babuschkin, raised $1.1 billion in a seed/Series A round led by Gener…
An unreleased Anthropic model significantly increased the lower bound of solutions for which the Riemann hypot…
Google Research and Google DeepMind have advanced AMIE, a research medical AI system built on Gemini and Proje…
ONESTRUCTION, Inc. built Ishigaki-IDS, a foundation model specialized for construction industry BIM workflows…
The AI news that matters, in one minute each morning.
Sign up free