AgentRQ
A self-hosted, real-time task manager for AI agents that lets a human control and review their own Claude, Codex or other ACP-compatible agents from mobile, web or desktop through the Model Context Protocol.
AgentRQ is a self-hosted platform for human-and-agent collaboration that uses the Model Context Protocol to let AI models such as Claude interact directly with a workspace's task-management system. Agents see workspace state through MCP, so they can pull their assigned tasks, update statuses, request permission for sensitive actions, and communicate with a human operator, synchronized in real time across devices.
Every task an agent creates appears on a visual list-or-Kanban board; tasks can be scheduled for a future launch date or a recurring cadence, checked by a background poller every minute; named Events such as qa_passed or deploy_finished can be fired by a completing task and wired, without code, to automatically create a new task in a subscribed workspace, with Workflows visualizing these event/workspace links as a graph. A task's History tab shows a lane-grouped timeline of every tool call and message in a run, and small on-device models handle auto-generated task titles and browser-based speech-to-text with no server call.
It is written in Go and released under the AGPL-3.0 license, and is designed to work with a user's own Claude subscription and any ACP-compatible harness.