Amazon, OpenAI, and Vercel unite on Agent Plugins — and Anthropic is missing
Curated by the Inblix editorial team
The fragmentation that’s plagued AI agent development just met a serious challenge. Amazon, Cursor, Microsoft, OpenAI, and Vercel have jointly released Agent Plugins, an open standard for packaging AI agent extensions so a single build works across their platforms.
The spec is refreshingly minimal. At its core, it’s a directory containing a plugin.json manifest file. Version 1.0.0 supports two component types: Agent Skills for reusable instructions and workflows, and MCP servers to connect agents with tools and data. That’s it — no marketplace rules, no permissions framework, no runtime constraints. The group is deliberately scoping this to packaging and discoverability only, developing the spec in the open on GitHub.
Here’s what makes this genuinely interesting rather than just another industry press release: the standard adopts the Model Context Protocol, which Anthropic created and open-sourced. MCP has become something of a lingua franca for connecting AI models to external tools, so its inclusion makes technical sense. But Anthropic itself is conspicuously absent from the roster of backers. The company recently shipped its own plugin system inside Cowork, its desktop agent tool, and maintains both MCP and Agent Skills as open standards. Sitting this one out raises the question of whether we’re watching cooperation or the start of a format war.
For developers who’ve been rebuilding agent extensions for each platform from scratch, a shared packaging standard solves a real headache. But the missing pieces — permissions, runtime behavior, marketplace distribution — are where the hard problems live. The spec’s narrow scope is either wise restraint or a sign that the real battles over agent interoperability haven’t started yet. Either way, when five major players agree on a common format, it puts pressure on everyone else to either join or justify going it alone.
💡 Key Takeaways
- Five major AI companies agreed on a single packaging format for agent extensions, eliminating the need to rebuild for each platform
- The standard adopts Anthropic's Model Context Protocol but launches without Anthropic's participation, hinting at a potential split in the ecosystem
- Version 1.0.0 intentionally sidesteps permissions, marketplaces, and runtime environments — the areas where real platform lock-in happens
Keep reading: See related articles below for more coverage on this topic.
Get smarter about AI
The sharpest AI news, curated daily. Delivered free to your inbox.