ACP standardizes communication between AI agents and code editors to enable interoperability and reduce editor-specific integrations. The protocol uses JSON-RPC over stdio for agent processes started by the code editor and reuses MCP (Model Context Protocol) specifications while adding custom types. Formatted text in ACP is based on Markdown. Libraries for TypeScript and Rust implement ACP. Google initiated the project because the Gemini CLI team sought deeper integration with Zed, making Zed the only editor with immediate support and Gemini CLI the reference agent implementation. ACP aims to create an ecosystem of agents analogous to the Language Server Protocol and to prevent VS Code lock-in.
ACP, which is still under development, is a standardized protocol for agent-editor communication. Agent processes are started by the code editor and communicate using JSON-RPC over stdio (standard input/output). ACP reuses MCP (Model Context Protocol) specifications where possible, while also adding its own custom types. Formatted text is based on Markdown. Libraries that implement ACP are provided for both TypeScript and Rust.
According to a post by Zed CEO and co-founder Nathan Sobo, the ACP project was initiated by Google, whose Gemini command-line interface (CLI) team were using Zed and wanted deeper integration. The consequence is that Zed is currently the only editor with immediate support, though there is also an extension for Neovim. The only ACP agent is Gemini CLI, which is the current reference implementation. This feature in Gemini CLI, which is open source, was originally called Zed integration schema upgrade.
Microsoft's VS Code is the most popular programmer's editor and IDE, and the company has loaded it with AI features, many of which are specific to GitHub Copilot. Some AI competitors have forked VS Code in order to integrate smoothly with their own offerings, but it is not an ideal solution, particularly as Microsoft controls the VS Code marketplace for extensions.
Collection
[
|
...
]