Human-in-the-Loop Agent Task Management for Claude Code
Documentation

AgentRQ Docs

Everything you need to keep humans in the loop while Claude Code works autonomously. From first setup to advanced MCP tool usage.

Quick Start

Go from zero to your first Claude Code task in under 60 seconds. Sign up, create a workspace, and paste one config snippet.

Read guide

Connect Claude Code

Configure the .mcp.json file, generate your MCP token, and wire up the notification channel in your CLAUDE.md.

Read guide

Workspaces

Understand workspaces — your isolated collaboration environment. Create one per project, invite agents, manage MCP tokens.

Read guide

Tasks

Learn the task lifecycle — how Claude creates tasks, you respond, and the agent continues. Handle approvals, rejections, and attachments.

Read guide

MCP Tools Reference

Complete reference for all 6 AgentRQ MCP tools: createTask, updateTaskStatus, reply, getWorkspace, getTaskMessages, downloadAttachment. Full parameter tables and code examples.

Read reference

How It All Fits Together

1
You create a workspace
Sign up at app.agentrq.com, create a workspace (one per project), and generate an MCP token.
2
Claude Code connects via MCP
Add AgentRQ to your .mcp.json. Claude Code picks it up automatically and gets access to all 6 MCP tools.
3
Claude creates tasks, you respond
When Claude needs input, it calls createTask. You get a real-time notification in the web UI. You reply, approve, or reject — Claude continues.
4
Everything streams in real-time
All task updates, replies, and status changes are pushed via Server-Sent Events — no polling, no delay. Claude listens via the notification channel.
Ready to start?
Free during beta — no credit card required.