Intermediate8 min
Agent Client Protocol (ACP): IDE Integration
ACP standardizes communication between coding agents and code editors — expose any Deep Agent as a coding assistant in Zed, JetBrains, VS Code, and Neovim via a single protocol.
Quick Reference
- →ACP = Agent Client Protocol — standardizes how coding agents talk to IDEs
- →deepagents-acp package exposes Deep Agents over ACP with one command
- →Supported editors: Zed, JetBrains (IntelliJ/PyCharm), VS Code (vscode-acp), Neovim
- →Stdio transport mode for local integration — editor launches the agent process
- →Distinct from MCP (tools) and A2A (agent-to-agent) — ACP is specifically for editor integration
- →Deep Agents CLI is itself an ACP-compatible coding agent
ACP vs. MCP vs. A2A
| Protocol | Purpose | Who Talks | Transport |
|---|---|---|---|
| MCP (Model Context Protocol) | Give agents access to tools and data | Agent ↔ Tool server | HTTP, stdio |
| A2A (Agent-to-Agent) | Agents communicate with each other | Agent ↔ Agent | HTTP (JSON-RPC) |
| ACP (Agent Client Protocol) | Agents integrate with code editors | Agent ↔ IDE | Stdio |
MCP extends what an agent can do (more tools). A2A lets agents collaborate. ACP puts the agent inside your IDE as a coding assistant. They're complementary — a coding agent can use MCP for tools, A2A for delegation, and ACP for the editor interface.