OpenAI, Amazon, Microsoft, Cursor, and Vercel launch Agent Plugins, an open standard for AI agent extensions

Vercel, together with Amazon Web Services, Anysphere (maker of Cursor), GitHub, Microsoft, and OpenAI, published Agent Plugins 1.0.0 on August 6, 2026 — an open, vendor-neutral packaging format meant to let a single AI agent extension work across competing agent products instead of being rebuilt for each one.

What's new

According to Vercel's announcement, Agent Plugins is "an open, vendor-neutral standard for plugins that extend AI agents," designed to give extension authors "one predictable, structured home" for the components they build instead of maintaining separate versions per client. A plugin is a directory containing a plugin.json manifest (at minimum a schema version and plugin name), an optional skills/ folder holding Agent Skills, and an optional mcp.json file configuring Model Context Protocol servers — bundling two previously separate extension mechanisms, MCP tool/data connections and reusable Agent Skills instructions, into one package format. The specification also allows optional client-specific namespaced directories for functionality that doesn't generalize across clients.

Vercel was explicit about the standard's scope: "The format is intentionally small and easy to implement, and it leaves installation, distribution, policy, user experience, and client-specific capabilities to each client." Governance sits with a Technical Steering Committee whose initial core maintainers are AWS, Cursor, Microsoft, OpenAI, and Vercel — Vercel initiated the proposal, and the other five companies refined it collaboratively rather than any single vendor controlling the roadmap.

At launch, ChatGPT, Codex, Cursor, GitHub Copilot, Kiro, and VS Code all support the format, giving Agent Plugins immediate reach across most of the widely used coding-agent surfaces.

Context

AI agent tooling has fragmented rapidly over the past two years: MCP emerged as a standard for connecting agents to external tools and data, while Agent Skills and similar mechanisms emerged separately for packaging reusable instruction sets. Extension authors targeting multiple agent products — ChatGPT, Cursor, VS Code, GitHub Copilot — have had to adapt the same underlying capability to each platform's own packaging conventions. Agent Plugins is an attempt to collapse that duplication into a single format that any compliant client can consume, similar in spirit to how earlier software ecosystems settled on shared package formats once enough vendors had independently reinvented the same wheel.

The timing is notable: OpenAI is highlighting the launch around GPT-5's one-year anniversary, using it to position itself as looking past single-model competition toward ecosystem-level interoperability — even though, per Vercel's own framing, this is explicitly not an OpenAI-led product.

Why it matters

A shared plugin format lowers the cost of building for the agent ecosystem rather than for one vendor's agent product, which should benefit smaller tool and skill developers most directly — they can now write once and reach ChatGPT, Cursor, Copilot, and VS Code users simultaneously rather than choosing which platform to prioritize. For the five founding companies, joint governance of a shared standard is also a hedge against fragmentation hardening into permanent lock-in around any single vendor's proprietary extension format, which would have made it harder for smaller players (and easier for the biggest ChatGPT-scale platform) to dominate distribution. Whether Agent Plugins actually sticks will depend on adoption beyond the launch partners — a standard with five co-authors and six launch clients is a strong start, but the test is whether third-party tool builders standardize on it rather than continuing to ship platform-specific integrations.