AgentRQ is a self-hosted, Apache-2.0 human-in-the-loop task manager that gives human operators and AI agents one shared workspace over the Model Context Protocol, aimed at people who run their own Claude subscriptions or any agent harness with ACP support.

What it is

AgentRQ is a platform for collaboration between human operators and AI agents, written in Go with a Vue.js 3 frontend and released under Apache-2.0. It lives in the agentic-AI ecosystem and speaks the Model Context Protocol through an integrated mcp-go SSE server that exposes tools and resources to models such as Claude. The architecture is decoupled and service-oriented: a Fiber-based REST API for workspace and task management, an MCP server, and a CoreMCP supervisor that is a global MCP server letting agents manage all workspaces, tasks and statistics across the platform. Storage is GORM over SQLite with user-scoped persistence, authentication is Google OAuth2 with JWT-based sessions, and an internal pub/sub event bus drives real-time SSE notifications. The frontend is built with Vue 3, Pinia and Tailwind CSS.

The concrete problem it solves is the gap between a person and the agent doing the work. It replaces polling loops and glue code: a background poller ticks every minute and spawns a scheduled task the instant it is due, so no server or agent has to stay awake and wait. It replaces YAML and decision-tree DSLs in release orchestration with a visible graph of events and workspaces. It also replaces the single browser tab chained to one agent session, because the board, tasks and conversations reach mobile, web and desktop.

Key capabilities

• Visual task board: every task Claude creates appears instantly, viewable on any device as a list or a Kanban, with a ⌘K (Ctrl+K off macOS) finder that matches any word in a task title or description from locally saved copy.
• Task scheduling: a launch date or recurring cadence of every 15 minutes, hourly, daily, weekly or custom days, spawned by a poller that runs each minute.
• Events: named signals such as qa_passed, deploy_finished and blog_published fired on completion; wiring an event to a workspace creates the next task automatically, with no polling and no glue code.
• Workflows: events and workspaces arranged on a graph, where dragging a workspace onto an event subscribes it and dragging an event onto a workspace emits it on completion.
• Tool Call History: the task detail view's History tab lays out a lane-grouped timeline of every tool call and message in a run across Input, Agent and Tools, searchable and showing whether each was allowed or denied.
• On-device privacy features: auto-title generation uses a small language model downloaded once and cached by the browser, and dictation uses an in-browser Whisper model, so neither text nor voice leaves the machine.
• Message send delay: a per-workspace countdown of 3s, 5s, 10s, 15s, 30s or 60s before a chat message reaches the agent, with Send Now and Cancel, off by default.

Who uses it and how

• Operators running their own Claude subscriptions, who let agents pull assigned tasks, update statuses and request permissions for sensitive actions over MCP.
• Teams with ACP-capable harnesses: the topics list acp-client, acp-gateway, claude-plugin, deepseek-harness-plugin, dsh-plugin and hermes-agent, so AgentRQ acts as the shared gateway those harnesses connect through.
• Release pipeline owners who model their process as a Workflow graph and chain tasks through events such as deploy_finished.
• Mobile-first operators who control agents from a phone, web browser or desktop client rather than a terminal session.
• Supervisors using the CoreMCP global server to manage workspaces, tasks and statistics across the whole platform from one agent.

Getting started

The README excerpt supplied here contains no install command, package name or Docker image; https://agentrq.com is listed as the homepage and is the reference point for hosted use. Self-hosting means running the Go Fiber API server and MCP server, building the Vue 3 frontend, and providing SQLite storage plus Google OAuth2 credentials.

How it compares

This registry provides no list of paid products that AgentRQ replaces, so no licence, cost or data-ownership comparison can be drawn from the supplied facts. It stands as an MCP and ACP gateway layer that sits alongside the harnesses named in its topics, rather than as a harness itself.

When to use it — and when not to

Choose AgentRQ when agents already speak MCP or ACP and the goal is one shared board for humans and agents across devices. A self-hoster must operate the Go backend, SQLite storage, a Google OAuth2 client and the frontend build, so teams wanting a turnkey hosted service, or teams with no MCP-capable agent, should look elsewhere. The README excerpt is detailed on features but silent on installation and deployment steps, which makes setup effort impossible to judge from the facts given.

curl -fsSL https://agentrq.com/install.sh | sh -s -- --quit