The Agent Client Protocol (ACP) standardizes the communication between code editors (IDEs, text editors, etc.) and coding agents—programs that use generative AI to autonomously modify code. While still under development, ACP is sufficiently advanced to enable the creation of compelling user experiences. Why ACP? Currently, AI coding agents and editors are tightly coupled, requiring each editor to implement custom integrations for every agent and vice versa. This leads to integration overhead, limited compatibility, and developer lock-in since choosing one agent often means accepting only its available interfaces. ACP addresses these issues by providing a standardized protocol for agent-editor communication, similar to the Language Server Protocol (LSP) for language server integration. This standardization allows agents to work with any compatible editor and editors to access a broader ecosystem of agents, fostering independent innovation and greater flexibility for developers. Overview ACP assumes the user is primarily working within their editor and wishes to use agents for specific tasks. Agents operate as sub-processes of the editor and communicate via JSON-RPC over standard input/output (stdio). The protocol reuses JSON representations from the MCP standard where possible and introduces custom types for agent-specific features like displaying diffs. User-readable text uses Markdown format to allow rich formatting without requiring HTML rendering capabilities from editors. Supported Editors Currently, ACP supports editors like: - Zed - Neovim (via the CodeCompanion plugin) Supported Agents Examples include: - Gemini (by Google) More agents and integrations are expected to be added in the future. In summary, ACP aims to decouple editors and AI coding agents via a common communication protocol, improving interoperability, reducing integration effort, and expanding user choice in tools.