---
title: "Pi ACP Setup | AgentRQ Docs"
url: https://agentrq.com/docs/agents/pi
description: "Pi ACP setup: run the Pi coding agent over the Agent Client Protocol and connect it to AgentRQ for human-in-the-loop task management."
---

[Docs](https://agentrq.com/docs/) / Agents / Integrations

Integrations

# Pi ACP Setup

Pi ACP setup: run the Pi coding agent over the Agent Client Protocol and connect it to AgentRQ for human-in-the-loop task management.

[Pi](https://github.com/badlogic/pi-mono) is a coding agent bridged to the Agent Client Protocol (ACP) via the `pi-acp` adapter, so it can run behind AgentRQ's gateway with no custom integration code.

Wondering which other coding agents support ACP? See the [ACP support table](https://agentrq.com/blog/which-coding-agents-support-acp-agent-client-protocol) for 33 agents.

Want Pi to use websites too? [Connect Pi to WebMCP](https://agentrq.com/docs/webmcp/pi) shows how it can call a site's WebMCP tools in your own Chrome.

## Does Pi Support ACP?

Yes. Pi publishes an ACP server to the [ACP registry](https://github.com/agentclientprotocol/registry) as `pi-acp`, so AgentRQ's gateway can fetch and run it by name — there is nothing to install by hand.

## What Connecting via ACP Gets You

Once Pi is bridged through AgentRQ's [ACP Gateway](https://agentrq.com/docs/connect-acp-agent), every task it works on shows up on your AgentRQ [task board](https://agentrq.com/features/task-board) in real time. You can approve risky commands, reply with extra context, or reassign work — from the dashboard or Slack — without touching the terminal Pi is running in.

## Connect Pi

Run both commands from your project root — the directory holding your `.mcp.json`.

Log in (the first run also downloads the agent):

bash

```bash
npx @agentrq/acp-gateway@latest --login --agent pi-acp
```

Start the bridge, and leave it running — it picks up tasks as you assign them:

bash

```bash
npx @agentrq/acp-gateway@latest --agent pi-acp
```

That `.mcp.json` is what tells the gateway which workspace to take tasks from:

json

```json
{
  "mcpServers": {
    "agentrq": {
      "type": "http",
      "url": "https://WORKSPACE_ID.mcp.agentrq.com/mcp?token=TOKEN"
    }
  }
}
```

See the [ACP Gateway guide](https://agentrq.com/docs/connect-acp-agent) for the full setup walkthrough, including where to find your Workspace ID and token.

## FAQ

**Does Pi work with AgentRQ?**

Yes — Pi connects via the `pi-acp` adapter, so it works with AgentRQ through the open-source [ACP Gateway](https://agentrq.com/docs/connect-acp-agent) with no custom integration code.

**How do I connect Pi to AgentRQ?**

Run `npx @agentrq/acp-gateway@latest --agent pi-acp` from your project root (check [pi-mono](https://github.com/badlogic/pi-mono) for any required flags), with a `.mcp.json` pointing at your AgentRQ workspace.

**What is the Agent Client Protocol (ACP)?**

ACP is an open, editor-agnostic protocol that lets any coding agent talk to any client over a standard JSON-RPC interface. AgentRQ's gateway implements the client side of ACP, so any ACP-compatible agent — including Pi — can use AgentRQ as its human-in-the-loop task manager.

## References

- → [Which coding agents support ACP?](https://agentrq.com/blog/which-coding-agents-support-acp-agent-client-protocol)
- → [pi-mono on GitHub](https://github.com/badlogic/pi-mono)
- → [Agent Client Protocol specification](https://agentclientprotocol.com)
- → [AgentRQ ACP Gateway guide](https://agentrq.com/docs/connect-acp-agent)
- → [Connect Pi to WebMCP](https://agentrq.com/docs/webmcp/pi)