Grok Build CLI: IDE Integration with ACP

@GrokInsider· Jun 6, 2026

Listen0:00 / 2:06

Grok is not only a terminal app. It speaks the Agent Client Protocol, an open JSON-RPC standard for AI agents, so it can run inside your editor with full session management, streamed thoughts, live tool visibility, and interactive permissions. If you would rather drive Grok from Zed or Neovim than a terminal, this is how.

New to Grok? Start at Grok Build CLI: Zero to Hero (Start Here).

Run Grok as an agent

The primary integration is stdio: Grok communicates over stdin and stdout in JSON-RPC. There is also a server mode for sharing one agent process across multiple clients over HTTP or WebSocket.

What ACP gives a client

Unlike headless mode, which runs one prompt and exits, agent mode keeps a persistent process and streams structured events. A client gets separate update types for response text, thinking tokens, tool calls, and plan entries, which is how editors render rich panels for thoughts and tool activity instead of a wall of text.

The protocol lifecycle is simple: initialize, create a session with a working directory, send prompts, receive streamed updates, and handle permission requests interactively.

Beyond the base protocol

Grok layers xAI-specific extension methods under an x.ai/ prefix for filesystem access, git and worktree operations, fuzzy search, terminal control, code navigation, session forking, and history rewind. There are dozens of them, which is what lets an editor offer deep integration rather than a chat box.

Compatible editors

ACP support is already shipping in real tools. Zed, Neovim (via CodeCompanion and avante.nvim), Emacs, and marimo notebooks all connect today, with JetBrains on the way. Official SDKs exist for TypeScript, Rust, Python, Go, and Kotlin if you want to build your own client.

If browsers cannot spawn local processes in your setup, run a WebSocket relay and have Grok connect out to it, which is how you build a web UI on top of the agent.

Continue the series

The full series