AgentRQ integrates with Claude Code via the native MCP protocol. One entry in .mcp.json gives Claude access to 7 purpose-built tools for
human
collaboration — no wrappers, no hacks, no polling loops.
{
"mcpServers": {
"agentrq": {
"type": "http",
"url": "https://WORKSPACE_ID.mcp.agentrq.com/mcp?token=TOKEN"
}
}
}Each tool covers one part of the Claude ↔ Human collaboration loop, with a short list of actions.
task
Claude's way to work with you. It opens a task with full context and files, pulls the next one from the queue, moves it from ongoing to completed or blocked, posts replies you see in the dashboard, asks you a question and waits for the answer, and fetches the files you attach.
memory
What this workspace remembers from earlier tasks. The memory belongs to the workspace, not the
agent, so a different model on a different machine starts with the same notes.
MEMORY.md is the index; detail lives in named memories
linked from it.
skills
The playbooks this workspace can use, its own and those shared into it. Search returns only names and descriptions, so having many skills stays cheap; load reads one file of a skill when the task needs it, and save writes back what the agent learned.
workspace
Reads the workspace's title and the mission you set, so Claude starts every session with the project's goals. publishEvent fires a named signal when a stage finishes, and every subscribed workspace gets its next task automatically.
siteTools
The websites you shared from the AgentRQ Chrome extension and the WebMCP tools they offer. Claude finds a tool, reads its definition, then runs it in your own Chrome, signed in as you. Anything the site does not mark read-only waits for your Allow or Deny in the task.
pano
Markdown docs shared by every workspace on your account: a release plan, a runbook, a handoff note. Agents write and read them, comment on them and attach files, so work in one workspace is visible from the next.
help
Explains everything else. With no payload it lists the tools; with a tool name it lists that
tool's actions; with task.create it prints the exact
payload and an example call.
task {"action": "create", "payload": {"title": "…", "body": "…"}},
and when it is unsure what an action needs, help tells it. Clients
you already installed keep working: the old tool names, such as createTask and loadMemory, still answer.
The most powerful way to use AgentRQ is to add instructions to your CLAUDE.md — telling Claude exactly when to ask you, what
context
to include, and what it can proceed with autonomously.
## AgentRQ — Human-in-the-Loop
At the start of every session:
→ Call `workspace get` to load mission context.
→ Report what you're connected to.
Before any of these, STOP and call `task create`:
- Database schema changes or migrations
- Anything that deletes or drops data
- Production deployments or config changes
- External API calls with write side effects
- Spending money (OpenAI, Stripe, etc.)
When creating a task, always include:
- What you've done so far
- Exactly what you need approval for
- The risk if they say yes
- The fallback if they say no
After human replies via channel:
→ Read the message carefully
→ Call `task updateStatus` with status "completed"
→ Continue with their instructions
# Be specific in tasks. Vague asks = slow unblock.No proprietary protocols. No vendor lock-in. AgentRQ uses standard MCP SSE transport — the same protocol powering the Claude ecosystem.