OpenAI, Vercel and rivals ship Agent Plugins 1.0 standard
A vendor-neutral format bundles Agent Skills and MCP server configs into one portable plugin, supported at launch by ChatGPT, Codex, Cursor, Copilot, Kiro and VS Code.
Six companies published Agent Plugins 1.0.0 on Thursday, an open specification for packaging agent extensions once and running them across competing agent clients.
The format
A plugin is a directory with a plugin.json manifest requiring only two fields — schema version and plugin name — plus fixed locations for its contents: skills live under skills/, MCP server configuration in mcp.json, with namespaced sections available for client-specific extras. The deliberate minimalism is the point: the directory layout, not a large schema, carries the contract, which keeps the barrier to implementing a reader low.
The specification bundles two things that until now travelled separately. Agent Skills are natural-language instruction documents that teach an agent a procedure; MCP servers expose tools and data. Extension authors have been repackaging the same components for each client's conventions. Agent Plugins gives them one artefact.
Who signed on
Vercel initiated the proposal, and engineers from Amazon Web Services, Anysphere (Cursor), GitHub, Microsoft, OpenAI and Vercel refined it to 1.0.0. The initial technical steering committee draws core maintainers from AWS, Cursor, Microsoft, OpenAI and Vercel. Clients supporting the format at launch are ChatGPT, Codex, Cursor, GitHub Copilot, Kiro and VS Code. Governance and the spec sit in a public repository, with documentation at agent-plugins.org.
One name is conspicuously missing from the launch roster: Anthropic, whose Agent Skills format the specification builds on and whose Claude Code is among the most widely used coding agents. Neither Anthropic nor Claude Code appears among the announced participants or launch clients; the company has not publicly commented on whether it intends to adopt the format.
Why it matters
MCP established a common wire protocol for tools; Agent Plugins is an attempt to standardise the layer above it — the distribution unit an author actually ships. If it holds, the practical effect is that a plugin ecosystem becomes portable rather than a per-vendor moat, and agent clients compete on execution quality instead of on which extensions they exclusively host. The unresolved question is whether a standard advanced by OpenAI, Microsoft, AWS and Vercel can become genuinely neutral without the vendor whose format supplied its starting point.