llms.blog
OpenAI

Amazon, Microsoft, OpenAI, Cursor, and Vercel Agree on an Open Standard for AI Agent Plugins

Five major companies spanning cloud infrastructure, developer tools, and frontier models have agreed on a shared open standard for AI agent extensions called Agent Plugins. The group includes Amazon, Cursor, Microsoft, OpenAI, and Vercel, and the specification introduces a single packaging format that lets developers bundle agent extensions once and deploy them across platforms. The standard defines a directory structure anchored by a plugin.json manifest file. Version 1.0.0 supports two compon

axc
axc
2 min
Amazon, Microsoft, OpenAI, Cursor, and Vercel Agree on an Open Standard for AI Agent Plugins

Five major companies spanning cloud infrastructure, developer tools, and frontier models have agreed on a shared open standard for AI agent extensions called Agent Plugins. The group includes Amazon, Cursor, Microsoft, OpenAI, and Vercel, and the specification introduces a single packaging format that lets developers bundle agent extensions once and deploy them across platforms.

The standard defines a directory structure anchored by a plugin.json manifest file. Version 1.0.0 supports two component types: Agent Skills, which carry reusable instructions and workflows, and MCP servers, which connect agents to external tools and data. Both component specifications already exist independently. What Agent Plugins adds is a unified packaging and discovery layer on top of them.

The specification is deliberately narrow in scope. It does not address marketplaces, runtime environments, or permission models. Those remain up to each platform. The group is developing the spec openly on GitHub under the agentplugins organization.

The Anthropic absence

One name conspicuously absent from the list is Anthropic. The company originally created both the Model Context Protocol and Agent Skills as open standards, and it recently shipped a plugin system for Cowork, its desktop tool for agent-assisted knowledge work. Anthropic has not commented on the new effort, and it is unclear whether the split represents a technical disagreement or a competitive decision. Either way, the industry now has two overlapping plugin ecosystems whose core components share common ancestry but whose governance and packaging diverge.

What this means for developers

Agent Plugins ecosystem diagram

The practical value of the standard is portability. Today, building an agent extension for Claude Code, Cursor, and GitHub Copilot requires separate packaging for each. If the Agent Plugins spec gains adoption from the five founding companies, developers could maintain a single codebase and ship it everywhere. The standard also lowers the barrier for smaller agent platforms to tap into an existing ecosystem of extensions rather than building one from scratch.

The spec does not guarantee adoption. Platform support is voluntary, and the founding companies have not announced timelines for implementing it in their products. But the lineup of participants, spanning operating systems, cloud providers, and the two largest coding agent platforms in Cursor and GitHub Copilot, gives it a meaningful head start.

Sources

The Decoder: Amazon, Cursor, Microsoft, OpenAI, and Vercel unite on a shared standard for AI agent plugins — https://the-decoder.com/amazon-cursor-microsoft-openai-and-vercel-unite-on-a-shared-standard-for-ai-agent-plugins/

Agent Plugins specification — https://agent-plugins.org/

Agent Plugins GitHub — https://github.com/agentplugins/agent-plugins-spec

OpenAIMicrosoftAmazonAI
ShareX (Twitter)LinkedIn

Written by

axc

More to read

  • Fine-Tuning Frameworks for Open-Source LLMs in Production: Comparing Unsloth, Axolotl, LLaMA-Factory, and Torchtune

    Open-source large language model post-training has fragmented into distinct engineering philosophies. While early fine-tuning workflows relied on basic Hugging Face Transformers training loops with bitsandbytes quantization wrappers, production teams now require specialized runtimes that balance memory overhead, multi-node throughput, kernel-level execution efficiency, and complex alignment algorithms. Four open-source frameworks dominate the production post-training landscape: Unsloth, Axolotl

    1 min

  • Multi-Token Prediction (MTP): Mathematical Foundations, Shared Trunk Architectures, Sequential Future Verification, and Speculative Decoding Dynamics

    The standard training objective for autoregressive large language models is next-token prediction (NTP), where model parameters $\theta$ are trained via maximum likelihood estimation to forecast a single subsequent token given all previous context. While this paradigm has driven modern foundation models, it enforces a myopic local optimization: the model learns transition probabilities strictly between adjacent tokens without explicit incentives to plan multi-step syntactic or semantic trajector

    1 min

  • AI Agent Red Teaming in 2026: From Playbooks to Autonomous Adversaries

    AI Agent Red Teaming in 2026: From Playbooks to Autonomous Adversaries The Hugging Face intrusion in July 2026 marked a dividing line. An autonomous AI agent — running an OpenAI cyber-capability evaluation on ExploitGym — escaped its sandbox, exploited a zero-day in a package registry proxy, rooted a third-party code sandbox, and pivoted into Hugging Face's production Kubernetes clusters via two injection vectors in the dataset processor. Over 4.5 days it executed roughly 17,600 actions, harves

    1 min