AgentRQ

What does this agent do, and when should you use it?

AgentRQ is a self-hostable human-agent collaboration and task-management platform built from a Go/Fiber backend, a Vue 3/Vite frontend, workspace-scoped MCP servers, and a global CoreMCP Supervisor. Human operators can break goals into tasks, assign work, and converse with agents, while agents pull assignments, update status, create tasks, download attachments, and reply through MCP. An internal pub/sub bus and SSE synchronize task changes, messages, and permission requests between agent sessions and the dashboard in real time. Persistent, user-scoped data is handled through GORM and SQLite, with Google OAuth2 and JWT-based sessions for authentication. Beyond its Claude Code channel integration, the repository documents gateways or extensions for ACP agents, OpenAI Codex, Gemini CLI, and DeepSeek Harness, plus a Docker self-hosting route.

Each workspace exposes its own MCP URL and token. An agent calls getWorkspace to read the workspace name, mission, and task statistics, then uses getTask either to fetch a task by ID or dequeue its next assigned notstarted task. It can create ordinary or recurring work with createTask and its optional cron_schedule, move work among notstarted, ongoing, blocked, and completed with updateTaskStatus, return live dashboard messages through reply, and retrieve files with downloadAttachment. Claude Code receives human interactions through notifications/claude/channel; for ACP clients without that native channel, @agentrq/acp-gateway reads .mcp.json, spawns the agent process, and relays assignments, messages, and permission requests between ACP and MCP. @agentrq/codex-gateway similarly bridges MCP to the Codex app-server protocol. At the account level, CoreMCP provides operations such as listWorkspaces, listAllTasks, updateTaskAssignee, respondToTask, and getAttachment, producing updated task, conversation, attachment, and workspace state.

1. A developer running several Claude Code projects can give each project a separate workspace MCP endpoint, allowing agents to claim tasks and report progress to one dashboard.
2. A team that requires human approval for sensitive actions can receive agent permission requests in task conversations and submit allow or deny decisions through Supervisor's respondToTask.
3. A program manager overseeing multiple projects can use CoreMCP to list tasks across workspaces, change assignments and priorities, and inspect workspace statistics.
4. A team using Gemini CLI or another ACP-compatible agent can place @agentrq/acp-gateway in front of the agent to deliver live task notifications to its active session.
5. An engineering group using OpenAI Codex can run @agentrq/codex-gateway to receive AgentRQ work while letting Codex reply and update task state through the project MCP configuration.
6. A team with recurring operational work can create tasks with cron_schedule and keep their execution state synchronized between agents and the web interface.

How do you install or deploy this agent?

Local development requires Go 1.21+, Node.js 18+ with npm, and an OAuth2 Client ID and Secret from Google Cloud Console. Put the credentials in backend/_config/base.yaml or backend/_config/development.yaml:

auth:
  google:
    client_id: "your-google-client-id"
    client_secret: "your-google-client-secret"

Install dependencies and start both application tiers from the repository root:

make install
make dev

The frontend is then available at http://localhost:5173. A pre-built Docker image is also documented as the production or development self-hosting route, although the supplied material delegates the exact Docker commands to SETUP.md. Install the ACP gateway when needed:

npm install -g @agentrq/acp-gateway

Install the Codex gateway when needed:

npm install -g @agentrq/codex-gateway@latest

How do you use this agent?

Obtain the workspace ID, complete MCP URL, and token from the workspace Setup modal. For Claude Code, create .mcp.json in the project root:

{
  "mcpServers": {
    "agentrq-WORKSPACE_ID": {
      "type": "http",
      "url": "YOUR_MCP_URL"
    }
  }
}

Create .claude/settings.local.json alongside it to enable the server and pre-approve the listed AgentRQ operations:

{
  "permissions": {
    "allow": [
      "mcp__agentrq-WORKSPACE_ID__updateTaskStatus",
      "mcp__agentrq-WORKSPACE_ID__getWorkspace",
      "mcp__agentrq-WORKSPACE_ID__reply",
      "mcp__agentrq-WORKSPACE_ID__createTask",
      "mcp__agentrq-WORKSPACE_ID__downloadAttachment",
      "mcp__agentrq-WORKSPACE_ID__getTask"
    ]
  },
  "enableAllProjectMcpServers": true,
  "enabledMcpjsonServers": ["agentrq-WORKSPACE_ID"]
}

Launch Claude Code from that project directory:

claude --dangerously-load-development-channels server:agentrq-WORKSPACE_ID

For Gemini CLI or another ACP agent, retain .mcp.json in the project root and invoke the agent through the gateway:

acp-gateway -- gemini --acp

Codex uses both .codex/config.toml for the agent's direct AgentRQ tool calls and .mcp.json for the gateway's task connection. A minimal gateway connection file is:

{
  "mcpServers": {
    "agentrq": {
      "type": "http",
      "url": "https://<WORKSPACEID>.mcp.agentrq.com/mcp?token=<TOKEN>"
    }
  }
}

Run the gateway from the workspace root containing that file:

codex-gateway

To provide the Codex command explicitly, use:

codex-gateway -- codex app-server

What are this agent's strengths and limitations?

How does this agent compare with similar options?

Key facts side by side with the most closely related agents.

How does FollowAgents rate this agent?

Recent releases

Latest entries from the project's GitHub Releases. Notes are quoted from upstream as written, not rewritten or tested by FollowAgents.

1. v0.13.0 2026-10-18
   Release notes excerpt

   What's Changed * feat(license): publish agentrqd and its installers under Apache-2.0 by @mustafaturan in https://github.com/agentrq/agentrq/pull/972 * feat(server): ship the server as one binary with the web app built in by @mustafaturan in https://github.com/agentrq/agentrq/pull/973 * fix(supervisor): clear context by default in the supervisor workspace by @mustafaturan in https://github.com/agentrq/agentrq/pull/975 * chore: bump minor version to v0.13.0 by @mustafaturan in https://github.com/agentrq/agentrq/pull/974 …

   View on GitHub ↗
2. v0.12.1 2026-10-18
   Release notes excerpt

   What's Changed * feat(acp): /logout logs a native ACP v1 agent out by @mustafaturan in https://github.com/agentrq/agentrq/pull/958 * fix(app): fit the Spin up panel on a phone screen by @mustafaturan in https://github.com/agentrq/agentrq/pull/959 * fix(acp): count a call as the workspace's own only under its exact server names by @mustafaturan in https://github.com/agentrq/agentrq/pull/960 * feat(acp): show a native ACP v1 agent's notices in the trajectory and as a toast by @mustafaturan in https://github.com/agentrq/agentrq/pull/961 …

   View on GitHub ↗
3. v0.12.0 2026-10-17
   Release notes excerpt

   What's Changed * test(repository): fail on SQL text built at run time by @mustafaturan in https://github.com/agentrq/agentrq/pull/941 * fix(acp): check ToolCall.name in workspaceTool for ACP v1 tool calls and permissions by @mustafaturan in https://github.com/agentrq/agentrq/pull/940 * fix(acp): return accept when elicitation/complete resolves a waiting URL elicitation by @mustafaturan in https://github.com/agentrq/agentrq/pull/942 …

   View on GitHub ↗

FAQ