AgentRQ

Name: AgentRQ
Author: agentrq

Task board, chat and approvals between people and coding agents, served to them over MCP

Official · agentrq Go Remote (hosted) · Streamable HTTP Productivity AI and ML Developer tools

GitHub npm Website

79 popularity score ★ 1.2k ↓ 1.7k/wk

Connect your client

Remote (hosted)

Add in one click

Add to Cursor Add to VS Code Add to LM Studio Add to Goose

Opens the app, which asks before it adds the server.

Or add it by command or config

Claude Code~/.mcp.json

{
  "mcpServers": {
    "agentrq": {
      "type": "http",
      "url": "https://mcp.agentrq.com/mcp"
    }
  }
}

Settings 10

`AGENTRQ_AUTH_JWT_SECRET`	Required: Self-hosted: secret for signing sessions, 32+ characters.
`AGENTRQ_AUTH_WORKSPACE_TOKEN_KEY`	Required: Self-hosted: exactly 32 bytes, encrypts workspace MCP tokens; changing it invalidates them.
`AGENTRQ_ACCOUNTS_OAUTH2_CLI_GOOGLE_CLIENT_ID`	Required: Self-hosted: Google OAuth2 client ID for sign-in.
`AGENTRQ_ACCOUNTS_OAUTH2_CLI_GOOGLE_CLIENT_SECRET`	Required: Self-hosted: Google OAuth2 client secret.
`AGENTRQ_BASE_URL`	Optional: Public address of the instance, e.g. http://localhost:2026.
`AGENTRQ_DOMAIN`	Optional: Domain workspace MCP subdomains hang off (localhost for local use).
`PORT`	Optional: Port the server listens on (2026 in the setup guide).
`AGENTRQ_POSTGRES_ENABLED`	Optional: Use PostgreSQL instead of the default SQLite file (false by default).
`AGENTRQ_SSL_ENABLED`	Optional: Built-in Let's Encrypt TLS; false by default.
`AGENTRQ_ACP_V1_ENABLED`	Optional: Turns on native ACP mode; off by default.

What it does

AgentRQ is a task manager where a person and their AI coding agents share a queue. You write tasks on a web, desktop or mobile board. An agent picks a task up through MCP, posts replies and asks before doing anything sensitive, and you answer from whatever device is at hand. You can use the hosted service at agentrq.com or run your own copy from the Docker image or a prebuilt binary.

There are two MCP servers in it. Each workspace gets its own endpoint, a URL with a token in it, and seven tools to work that workspace's queue, notes and skills plus the shared docs. The account-wide Supervisor at https://mcp.agentrq.com/mcp signs in with OAuth. Its nine tools manage workspaces, tasks in any workspace, events and workflows, enrolled machines and the Pano docs, and it also offers five resources and three prompts. Each tool takes an `action` and a `payload`, and you can ask `help` for any action's fields with an example.

Since v0.14.1 a supervising agent can pull just a task's conversation with `task.messages`. A client on MCP revision 2026-07-28 can subscribe to task resources to hear when a status changes or an agent wants permission, instead of polling. List resources hold a workspace's 50 newest tasks (task_resources.go). There is also a YOLO switch per task: while it is on, permission requests from that agent are allowed straight away and logged as auto-allowed. From the Supervisor it is `task.updateAllowAll`.

Claude Code gets live notifications through its channel feature. Codex, Gemini and others get them through a separate @agentrq/acp-gateway bridge, which needs native ACP v1 (on by default in base.yaml). Plugins for Claude Code and Cursor ship from this repository too. On npm, @agentrq/agentrq-ws is a command-line client for workspace tools, not the server.

Old one-name-per-action tools (createTask, listWorkspaces and so on) still work on both servers and get mapped onto action tools, so setups written for earlier releases need no changes.

What it needs

To run it yourself	DockerGo 1.27.1+Node.js 18+
AgentRQ	FreeHosted at agentrq.com; workspace URL and token from the workspace's Setup dialog, OAuth sign-in for the Supervisor
AgentRQ (self-hosted)	Your own instanceNeeds a Google (or GitHub) OAuth2 client ID and secret for sign-in
Also	Node.js 22+ for the ACP gateway used by Codex, Antigravity and other non-Claude agents

Hosted: copy a workspace's MCP URL from its Setup dialog, or connect the Supervisor and finish the OAuth sign-in. In Cursor, install the AgentRQ plugin from Customize; the workspace plugin takes its MCP URL under Plugins, Configure. Self-hosted: Docker (or Go 1.27.1+ and Node.js 18+ to build from source), a Google or GitHub OAuth2 client ID and secret, and a JWT secret of 32+ characters. Start it with `docker run -d -p 2026:2026 --env-file .env -v <ABSOLUTE_PATH>:/_storage agentrq/agentrq:latest`. Live notifications need Claude Code started with `--dangerously-load-development-channels`, or `npx -y @agentrq/acp-gateway@latest --agent <AGENT>` (Node 22+) for other agents.

Limits

Not covered A workspace endpoint sees only its own workspace; working across workspaces needs the Supervisor
Read only The Supervisor's memory and skills tools only list, search and read; it cannot change them or publish events
Not covered Machines cannot be enrolled remotely: createEnrolmentCode returns a command that must be run on the machine itself
Limit Live notifications reach Claude Code natively; other agents need the ACP gateway and a server with ACP v1 left on
Limit Task resource subscriptions need a client on MCP revision 2026-07-28 and cover the caller's own workspaces, at most 100 per listen
Size cap The workspace and delegated task resources return the 50 newest tasks only
Limit Changing AGENTRQ_AUTH_WORKSPACE_TOKEN_KEY makes every existing workspace token unreadable
Limit If AGENTRQ_AUTH_WORKSPACE_TOKEN_KEY is unset the code falls back to a fixed placeholder key published in base.yaml
Platform Desktop builds are unsigned, so a downloaded build warns on first launch

Each workspace endpoint sees only its own workspace, and the Supervisor reads memory and skills without changing them. Native ACP mode is on by default. A server that sets AGENTRQ_ACP_V1_ENABLED=false answers 404 on those endpoints, so the ACP gateway used by Codex and other non-Claude agents cannot connect to it. Changing AGENTRQ_AUTH_WORKSPACE_TOKEN_KEY makes every existing workspace token unreadable.

Who it suits

Suits a developer who runs Claude Code, Codex or similar agents on their own machines and wants to hand them tasks, approve their actions and reply to them from a phone or browser instead of a terminal, with the option to keep all data on a self-hosted instance. It is a whole platform rather than a thin connector: self-hosting means running a web service with Google or GitHub sign-in, so someone who only wants an agent to read a list of to-dos may prefer a simpler task-list server. Agents other than Claude Code need the ACP gateway and a server with ACP v1 left on. The Supervisor can launch agents on enrolled machines, so give it only to an agent you trust with that.

Tools 10

1 only reads, 3 change things, 6 can delete or overwrite. A ? marks an optional parameter.

task deletes: Workspace: create, dequeue, update and reply to tasks, fetch attachments and ask the human. Supervisor: list, read messages, answer permissions, toggle YOLO and delete across workspacesaction: enumpayload?: object
memory deletes: Workspace: load, save or delete long-lived notes; Supervisor: list and read them onlyaction: enumpayload?: object
skills deletes: Workspace: search, load, save or delete SKILL.md playbooks; Supervisor: search and read them onlyaction: enumpayload?: object
workspace changes: Workspace: read its mission and publish an event; Supervisor: list, create, update, fork, merge forks and get statsaction: enumpayload?: object
siteTools changes: Find and call WebMCP tools of websites the human shared from the Chrome extension, run in their signed-in Chromeaction: enumpayload?: object
pano deletes: Save, read, comment on, attach files to and delete the account's shared markdown docsaction: enumpayload?: object
events deletes: Supervisor: define named events and the triggers that create tasks when they fireaction: enumpayload?: object
workflows deletes: Supervisor: build and edit graphs of events and steps, also as an indented text documentaction: enumpayload?: object
machines changes: Supervisor: make enrolment codes, list enrolled machines, and start or stop agents on themaction: enumpayload?: object
help: Print every tool's actions, one tool's actions, or one action's payload with an exampleaction?: string

Resources 5

agentrq://guides/new-workspace: Supervisor: guide to setting up a new workspace for an agent
agentrq://guides/agentrqd-setup: Supervisor: guide to installing agentrqd and enrolling a machine
agentrq://workspaces/{workspaceId}/tasks/{taskId}: Supervisor: one task's status, whether it awaits a reply, and its last five text messages
agentrq://workspaces/{workspaceId}/tasks: Supervisor: the 50 newest tasks of a workspace, subscribable for changes
agentrq://tasks/delegated-by/{workspaceId}: Supervisor: the 50 newest tasks a workspace created elsewhere, subscribable for changes

Prompts 3

new-workspace: Supervisor: create a workspace for a stated purpose and list what a human must still connect
setup-agentrqd: Supervisor: make an enrolment code and give the install and enrol commands for a new machine
workspace-status: Supervisor: summarise what is happening across every workspace

Popularity

79 popularity score, on our list (the bar is 36)

★ 1.2k ↓ 1.7k/wk

GitHub stars	1,179	36.9 points
Stars in 30 days	+40	8.1 points
npm downloads, last week	1,737	25.9 points
Official server	Yes	8 points

Measured 2026-11-07. How the score works.

Where it is published

Code	agentrq/agentrq ★ 1.2k
npm	@agentrq/agentrq-ws
Docker Hub	agentrq/agentrq
Hosted	`https://mcp.agentrq.com/mcp` Streamable HTTP · OAuth sign-in

What it can reach

Runs	On the maker's servers (remote)
Reaches	Your account at a service, Sends messages, Drives a browser, Runs shell commands
Changes things	Yes: it can write, send, create or delete
Sign-in	OAuth sign-in
Licence	The maker's terms (a hosted service)
Last code change	2026-11-05

A workspace endpoint is authorised by a token in its URL query string, so the URL itself is the secret. The Supervisor (also through the Cursor and Claude Code plugins) signs in with OAuth. Tools create, change and delete tasks, notes, skills and shared docs. With YOLO on for a task (the Supervisor's task.updateAllowAll, or a workspace or trigger default) the agent's permission requests are allowed without a person answering. siteTools.call runs a shared website's WebMCP tools in the person's signed-in Chrome, and a tool the site does not mark read-only waits for approval. The machines tool starts agents on machines enrolled with agentrqd, which run commands there as the daemon's user. Hosted data sits on agentrq.com. Self-hosted, it stays in your SQLite or PostgreSQL database.

Over time

GitHub stars 1.2k +40 (+4 %)

1.2k1.1k

8 Oct 20267 Nov 2026

npm downloads a week 1.7k +259 (+18 %)

1.7k598

28 Sept 20267 Nov 2026

Before the site measured it, read from npm's daily download counts.

Popularity score 79 +17 (+27 %)

7945

8 Oct 20267 Nov 2026

What is rising across the directory

Sources

Page checked 2026-11-02.

Badge for your README

Shows this server's rank in its category, and stars for a server that runs on your machine. It updates every hour.

MarkdownREADME.md

[![AgentRQ on EveryMCPServer](https://everymcpserver.com/badge/agentrq.svg)](https://everymcpserver.com/servers/agentrq)

Connect your client

Add in one click

Or add it by command or config

Settings 10

What it does

What it needs

Limits

Who it suits

Tools 10

Resources 5

Prompts 3

Similar servers

Agentic Flow

agent-harnesses MCP server

agentistics

Ruflo

Agent Network

Agent Harness Kit