← Back to glossary
+Suggest a term
Concept·Agents & Automation·Added today

Agent Client Protocol

Also known as: ACP, ACP protocol, editor-to-agent protocol

An open protocol that standardizes how code editors talk to coding agents. Created by Zed Industries, it lets any ACP-compatible agent run inside any ACP-compatible editor. Think of it as the LSP for AI coding agents: write once, plug in anywhere.

The Agent Client Protocol (ACP) was created by Zed Industries in August 2025 and defines a standard communication layer between code editors and the AI coding agents that work inside them. Before ACP, every agent was tightly coupled to one editor. With ACP, an agent that speaks the protocol can run inside any editor that also speaks it, and editors can host multiple competing agents without custom integration work for each one.

ACP runs as JSON-RPC 2.0 over stdin/stdout, which keeps it lightweight and easy to implement. It handles the flow of context from editor to agent: what files are open, what the cursor sees, where the diff should land, when to ask for approval. A useful mental model is to compare it to LSP, the Language Server Protocol, which solved the same problem for language tooling. Before LSP, every language had to build separate integrations for every editor. ACP attempts the same fix for agents.

By June 2026, ACP had been adopted by Zed natively, JetBrains IDEs, GitHub, Google, and more than 25 AI agents, but it broke into wider builder awareness when Cognition relaunched Windsurf as Devin Desktop with ACP support, letting Codex, Claude Agent, OpenCode, and custom in-house agents all run inside a single editor surface. Builders building their own agents now often ask whether their agent speaks ACP as a baseline interoperability question.

This definition is AI-generated and refreshed weekly. It may contain inaccuracies. Use your own judgment, especially for production decisions.
Related terms
MCPAgentic IDEDevin DesktopA2A protocolCascade