Which AI coding tools support Agent Plugins today?

Agent Plugins is read by six clients at launch: ChatGPT, Codex, Cursor, GitHub Copilot, Kiro and VS Code. Each client keeps control of how a plugin is installed, distributed and surfaced to users, so support means the client can load the same folder, not that every client behaves identically once it does.

Is Google a maintainer of Agent Plugins?

Google announced on August 6, 2026 that it joined Agent Plugins as a core maintainer, represented on the Technical Steering Committee by Kevin Hou of Google DeepMind. Google already ships two plugins in the format: its Agents CLI, which carries skills for building and deploying agents, and the Data Agent Kit for BigQuery, Spanner and Cloud SQL.

How is Agent Plugins different from MCP?

MCP defines how an agent talks to an external tool or data source. Agent Plugins is a layer above that: it defines the folder you ship, bundling MCP server declarations in mcp.json together with Agent Skills under skills/. Agent Plugins does not replace MCP, it packages MCP servers and skills together so both travel as one unit.

Can a plugin still carry client-specific settings?

Yes. Agent Plugins reserves reverse-domain namespaces, such as com.example.client/, for anything one client needs and others do not. The portable parts stay in the fixed locations every client understands, so a plugin can hold extras for one tool without breaking in the rest.

Agent Plugins · 2026-08-06 · major

Agent Plugins 1.0.0 — one plugin format across Cursor, Copilot and ChatGPT

Agent Plugins 1.0.0 is an open standard for packaging agent skills and MCP servers into one portable folder. Amazon, Cursor, Microsoft, OpenAI and Vercel maintain it, and ChatGPT, Codex, GitHub Copilot, Kiro and VS Code already read it.

One directory format packages agent skills and MCP servers so the same folder runs in any supporting client.

Key specs

Spec repo stars	659

Quick facts

Spec version	1.0.0
Released	August 6, 2026
Core maintainers	Amazon, Cursor, Microsoft, OpenAI, Vercel
Supported clients	ChatGPT, Codex, Cursor, GitHub Copilot, Kiro, VS Code
Package layout	plugin.json + skills/ + mcp.json
License	CC-BY-4.0 (spec), Apache-2.0 (schemas and code)
Governance	Community-governed, run by a Technical Steering Committee

What is it?

Agent Plugins 1.0.0 fixes a single directory format for shipping agent extensions: a plugin.json manifest at the root, a skills/ folder, and an mcp.json file. Vercel published the release on August 6, 2026, with Amazon, Cursor, Microsoft and OpenAI alongside it on the steering committee. The same folder can be copied into any client that supports the spec.

How does it work?

Each skill sits in its own subdirectory under skills/, using the SKILL.md layout the Agent Skills specification already defines, with optional scripts/ and references/ folders next to it. MCP servers are declared in mcp.json, and every entry carries an explicit type. Anything a single client needs goes in a reverse-domain namespace, which keeps the portable parts of a plugin in places every client knows how to find.

Why does it matter?

Anyone building agent extensions has had to maintain a separate package for each tool that might load it. A plugin written to this spec covers six clients at once, so the duplicate codebases go away. The spec is deliberately small: it says nothing about plugin stores or security, and each client still decides installation, distribution and policy for itself.

Who is it for?

developers who ship agent skills and MCP servers

Frequently asked questions

Which AI coding tools support Agent Plugins today?: Agent Plugins is read by six clients at launch: ChatGPT, Codex, Cursor, GitHub Copilot, Kiro and VS Code. Each client keeps control of how a plugin is installed, distributed and surfaced to users, so support means the client can load the same folder, not that every client behaves identically once it does.
Is Google a maintainer of Agent Plugins?: Google announced on August 6, 2026 that it joined Agent Plugins as a core maintainer, represented on the Technical Steering Committee by Kevin Hou of Google DeepMind. Google already ships two plugins in the format: its Agents CLI, which carries skills for building and deploying agents, and the Data Agent Kit for BigQuery, Spanner and Cloud SQL.
How is Agent Plugins different from MCP?: MCP defines how an agent talks to an external tool or data source. Agent Plugins is a layer above that: it defines the folder you ship, bundling MCP server declarations in mcp.json together with Agent Skills under skills/. Agent Plugins does not replace MCP, it packages MCP servers and skills together so both travel as one unit.
Can a plugin still carry client-specific settings?: Yes. Agent Plugins reserves reverse-domain namespaces, such as com.example.client/, for anything one client needs and others do not. The portable parts stay in the fixed locations every client understands, so a plugin can hold extras for one tool without breaking in the rest.

Try it

https://github.com/agentplugins/agent-plugins-spec