Introduction

The Agent Client Protocol (ACP) is a standardized protocol designed to facilitate communication between code editors or integrated development environments (IDEs) and coding agents. Its primary aim is to address the challenges of interoperability between editors and agents, which traditionally require custom integrations for each pairing. By introducing a common protocol, ACP enables agents and editors to work together without the need for editor-specific APIs or agent-specific interfaces, reducing integration overhead and broadening compatibility.

This protocol is intended for scenarios where developers primarily use their code editor and wish to leverage external agents to assist with coding tasks. ACP supports both local and remote agent configurations. Local agents operate as sub-processes of the editor and communicate using JSON-RPC over standard input and output streams. For remote agents, communication occurs over HTTP or WebSocket, although full support for remote agents is still under development. The protocol reuses JSON representations from MCP where applicable and introduces custom types tailored to agentic coding user experiences, such as the ability to display code diffs. Markdown is the default format for user-readable text, chosen for its flexibility in representing rich formatting without requiring editors to render HTML.

ACP decouples the development of editors and agents, allowing each to innovate independently. Editors that implement ACP gain access to a broader ecosystem of agents, while agents that support ACP become compatible with any editor that adopts the protocol. This approach aims to give developers greater freedom in choosing their tools and to minimize lock-in to specific platforms or interfaces. The protocol is being developed in collaboration with agentic platforms to ensure it meets the requirements of both local and cloud-hosted deployment scenarios.