Human-in-the-Loop Agent Task Management for Claude, Gemini & ACP Agents
Back to Blog
April 18, 2026 | AgentRQ Team

AgentRQ + ACP: Connecting Gemini CLI and Beyond to Your Remote Task Channel

The AI agent ecosystem is standardizing fast. With the rise of the Agent Client Protocol (ACP), tools like Gemini CLI, Claude Code, and Zed are now speaking a common language.

But there has always been a "last mile" problem: most of these agents are designed to run locally in your terminal. When you step away from your machine, your agent's productivity stops.

Today, we're bridging that gap with the AgentRQ ACP Gateway.

This new open-source tool allows you to connect any ACP-compatible agent—starting with Gemini CLI—to your AgentRQ workspaces. The result? You can now delegate tasks to Gemini from your phone, monitor its progress from a browser, and provide human-in-the-loop feedback from anywhere in the world.

What is the AgentRQ ACP Gateway?

The ACP Gateway is a bridge between two worlds.

On one side, it speaks ACP, the standard protocol for AI coding agents. On the other side, it connects to AgentRQ via MCP (Model Context Protocol).

When you run an agent like Gemini CLI through the gateway, it essentially "promotes" your local terminal agent to a remote-capable collaborator. The gateway listens for new tasks arriving in your AgentRQ workspace and forwards them to Gemini. When Gemini needs to ask a question or request permission, the gateway surfaces that back to you via the AgentRQ UI.

Why This is a Game-Changer for Gemini CLI Users

Gemini 2.0 and its successors have brought incredible reasoning capabilities to the terminal. But until now, using Gemini CLI meant being tethered to your desk.

By connecting Gemini CLI to AgentRQ via the ACP Gateway, you unlock:

1. Real-Time Remote Collaboration

You remember a task while on the train. Instead of waiting until you get home, you open the AgentRQ mobile app, create a task, and assign it to your "Gemini-MacBook" workspace. Back at home, your local Gemini instance picks up the task immediately and starts working.

2. Human-in-the-Loop Everywhere

Gemini is powerful, but sometimes it needs your judgment. Should it delete that old config file? Should it use a specific library? When Gemini hits a decision point, it creates a "Task Message" in AgentRQ. You get a notification, review the context, and reply with "Yes" or "No"—all without touching a terminal.

3. Headless "Always-On" Productivity

The ACP Gateway includes a smart polling loop. If you leave it running, it will continuously check for the next pending task in your workspace. You can queue up a dozen refactoring tasks before you go to sleep, and wake up to a series of completed pull requests and status updates.

How to Connect Gemini CLI to AgentRQ

Getting started is simple. You just need Gemini CLI installed and an AgentRQ account.

Step 1: Install the Gateway

You can run it directly via npx:

bash 
npx @agentrq/acp-gateway -- gemini --acp

Step 2: Configure your .mcp.json

The gateway looks for a .mcp.json file to know which AgentRQ workspace to connect to. You can find your workspace URL and token in the AgentRQ Settings.

json 
{
  "mcpServers": {
    "agentrq": {
      "type": "http",
      "url": "https://YOUR_WORKSPACE_ID.mcp.agentrq.com/mcp?token=YOUR_TOKEN"
    }
  }
}

Step 3: Start Working

Once connected, Gemini will show up as an active agent in your dashboard. You can now start sending tasks.

Beyond Gemini: The ACP Standard

While we're excited about the Gemini CLI integration, the beauty of the Agent Client Protocol is that it is an open standard. Any agent that implements ACP can now be bridged to AgentRQ using this gateway.

Whether you're using Claude Code, a custom internal agent, or the next generation of ACP-compatible IDEs like Zed or JetBrains, the AgentRQ ACP Gateway provides the persistent, remote-accessible communication layer that turns a "local tool" into a "global team member."

Unlock Your Agent's Potential

The future of software engineering isn't just about better models; it's about better workflows. By decoupling the agent's execution (local machine) from the human's supervision (anywhere), we're making it possible to scale your productivity like never before.

Ready to get started?

---

FAQ

Can I use this with Claude Code?

Yes! While Claude Code has native support for AgentRQ's notification channels, you can also run it through the ACP Gateway if you prefer the ACP-standardized interaction model. See our Claude Code guide for more details.

Does this work with private codebases?

Absolutely. Your agent runs locally on your machine. Only the task descriptions and your feedback travel through AgentRQ's encrypted channels. Your source code never leaves your local environment unless your agent specifically interacts with a remote tool you've configured.

What other agents support ACP?

The list is growing every day. In addition to Gemini CLI and Claude Code, many experimental agents and newer IDE integrations are adopting the standard. Check the official ACP registry for the latest updates.

Start Free