
AgentRQ Supervisor MCP Server: Setup for Claude Code, Cursor and Other Apps
Lets Claude, Cursor and other assistants manage AgentRQ workspaces, tasks, workflows and the agents that work on them.
AgentRQ is a task board where people and AI agents work together, and this is its own supervisor server. It lets an assistant list and create tasks and workspaces, reply to agents, build workflows, and start or stop agents on your machines. You sign in with your AgentRQ account in the browser. Know before you use it that it can answer an agent's permission requests and start agents that run code on your computers.
Good for
- People who run several AI agents and want one assistant to hand out and track their tasks
- Teams building event-driven agent workflows across workspaces
- Developers who already use AgentRQ and want their assistant to supervise from Claude Code or Cursor
- Anyone who wants to start and stop coding agents on their own machines from chat
Works in
14 should work
Tap an app to see how to set it up and any known problems.
Supports hosted MCP servers.
Supports hosted MCP servers.
Supports hosted MCP servers.
Supports hosted MCP servers.
Supports hosted MCP servers.
Supports hosted MCP servers.
Supports hosted MCP servers.
Supports hosted MCP servers.
Supports hosted MCP servers.
Supports hosted MCP servers.
Supports hosted MCP servers.
Supports hosted MCP servers.
Supports hosted MCP servers.
Supports hosted MCP servers.
Install AgentRQ MCP
Claude Code
- 1
Copy this prompt.
Add the AgentRQ MCP server to Claude Code. Run this command: /plugin marketplace add https://github.com/agentrq/agentrq /plugin install agentrq@agentrq Then check that it loaded and tell me if I need to sign in or restart.
- 2
Paste it into Claude Code. It shows the change and asks before running it.
Requires AgentRQ account.
The plugin method comes from the README and also adds an AgentRQ skill. The README's own example puts the same address in ~/.mcp.json as an http server. The plugin named agentrq-workspace is a different server for one workspace. A browser link appears the first time so you can sign in.
From the project README. Source.
Tools
Lists, creates, replies to and manages tasks across workspaces
One tool with many actions: list or search tasks in one workspace or all, create and read them, reply, change status, assignee, order or schedule, answer a permission request, set allow-all, fetch attachments, and delete a task with its messages.
Try: “Show every blocked task across my workspaces and reply to the one about the migration.”
Delete ·
taskLists, creates, updates and forks workspaces
Lists your workspaces and their agents, reads one, creates one, updates its settings, shows 7 or 30 day stats, forks it, and merges a fork back and removes it. It cannot delete a workspace, but a merge can delete the fork's folder on the machine if you ask.
Try: “Fork the api workspace so a second agent can take the test tasks.”
Delete ·
workspaceLists your computers and starts or stops agents on them
Lists machines that run the agentrqd helper, makes a one-time code to enrol a new machine, starts a Claude Code or ACP gateway agent in a workspace's folder on a machine, and stops it.
Try: “Start a Claude Code agent for the docs workspace on dev-box.”
Runs code ·
machinesManages named signals and what they trigger
Creates, reads, changes and deletes events, and the triggers that create a task in a workspace when an event fires. It also lists tasks an event has started. Events are published by the agents, not from here.
Try: “Create a deploy_finished event that gives the release workspace a task.”
Delete ·
eventsBuilds and edits workflow graphs
Lists, creates, changes and deletes workflows and their steps, and reads or replaces a whole graph as indented text. It also lists the tasks a workflow started.
Try: “Rewrite the release workflow so qa_passed triggers the deploy workspace.”
Delete ·
workflowsReads what a workspace's agent remembers
Read only. Lists a workspace's memory notes with size and change time, and reads one in full, starting from the memory.md index.
Try: “What does the web workspace remember about our test conventions?”
Network ·
memorySearches and reads the account's skills
Read only. Searches skills by name or description, with whether each is on in a workspace, and reads one file by its skill:// address.
Try: “Which skills does the docs workspace have turned on?”
Network ·
skillsReads and writes the shared markdown docs
Saves, reads, comments on, attaches files to and deletes the account's shared markdown docs (the Pano) that every workspace can see.
Try: “Save a Pano doc with today's release checklist.”
Delete ·
panoUsed by the assistant behind the scenes
Explains how to call the other tools
Read only. With no input it lists every tool and action. Given a tool or tool.action it prints the fields and an example.
Try: “Show me the fields for creating a task.”
Read ·
help
Read from the source code
Permissions
high risk · 4 things to watch
High risk
The assistant can delete tasks, events, workflows and docs, approve an agent's permission requests, and start coding agents on your machines. Give it to an assistant only when you want that reach, and check each permission request it answers.
What it can do
- Read your data
- Create and change things
- Delete things
- Read-only mode
Watch out for
It can start agents that run code on your machines
The machines tool starts a Claude Code or gateway agent in a workspace folder on a computer you enrolled, and stops it mid-task. Only allow it where you want agents running on those machines. github.com
It can approve an agent's permission requests
The task tool can allow or deny a permission request and set a task to allow all commands. That removes the human check the platform is built around, so review what it approves. github.com
It can delete tasks, events, workflows and docs
Deleting a task also deletes its messages and attachments. A fork merge can delete the fork's folder on the machine when asked. Back up anything you cannot recreate. github.com
Workspace servers use a token in the address
Each workspace also has its own server address with a token in the URL. That is a separate server for single-workspace agents. Treat that address as a secret. github.com
Security advisories
No published security advisories as of Nov 1, 2026.
Technical details
| Turning on read-only | No read-only mode is documented. Only the memory, skills and help tools are read-only. |
|---|---|
| Talks to | mcp.agentrq.com (hosted by AgentRQ) |
| Sign-in and keys | Sign-in with your AgentRQ account through your browser (OAuth2). The tokens are kept by your AI app. |
Questions
Is AgentRQ official?
Yes. This is the Supervisor MCP server of the AgentRQ platform, published from the agentrq/agentrq repo. The code is AGPL-3.0 licensed.
What is the AgentRQ MCP server URL?
https://mcp.agentrq.com/mcp. It signs you in with OAuth. Each workspace also has its own address with a token, which is a different server for single-workspace agents.
Supervisor or workspace server: which do I need?
Use the Supervisor to oversee all your workspaces from one assistant. Use a workspace address for an agent that should work one workspace's queue only. The Supervisor is the one covered here.
Does it need an AgentRQ account?
Yes. You sign in with your account in the browser the first time. You can also self-host AgentRQ; the Claude Code plugin for self-hosted servers asks for your own server's Supervisor address.
Can it run read-only?
No. No read-only mode is documented. Only the memory, skills and help tools only read, and the rest can change or delete things.
Does it work in Claude Desktop or ChatGPT?
AgentRQ documents Claude Code, Cursor and Gemini CLI. It does not document other apps, so a connector there may not work.
How do I add it to Claude Code?
Use the install box above. AgentRQ offers a plugin from its own marketplace, or you can add the hosted address as an http server. A browser link appears the first time to sign in.
Can it delete a workspace?
No. The tool description says workspaces cannot be deleted here. It can delete tasks, events, workflows and docs, and merge a fork back into its parent.
Information
updated Nov 2, 2026
- Publisher
- Official
- Endpoint
https://mcp.agentrq.com/mcp- Transport
- Streamable HTTP
- Auth
- OAuth
- Version
- 0.15.1
- License
- AGPL-3.0
- Last commit
- Oct 31, 2026
- Open issues
- 0
- Listed in
- Official MCP registry
Sources
Docs, issue trackers and discussions this page is built from. Issues, quotes and findings above link to their own source.
Show all 3Hide
- github.com · github.com
- github.com · github.com
- github.com · github.com