Tutorial 10 min2026-06-08

Your First Guided Routine in 10 Minutes

Connect Claude to CloudClawer, set a daily schedule, drop in a task, and watch the agent brief you and work — with cost tracking from the first token.

What you'll set up

MCP connection so Claude can talk to CloudClawer

A daily schedule with one time slot

A guided task that fires into that slot

Cost and fire-log visibility so you know exactly what ran

A Guided Routine is how you put Claude on a schedule. You set a time slot, drop tasks into a queue, and at the scheduled time CloudClawer fires Claude with your tasks injected automatically — no copy-pasting context, no manual trigger. By the end of this guide your agent will be briefing you on daily work items on its own.

Prerequisites: A CloudClawer account (free tier works) and Claude.ai or the Claude desktop app. No CLI setup needed for the basic path.

1

Connect the MCP server

CloudClawer exposes its tools to Claude via MCP. There are two paths — pick whichever fits.

Path A — Claude.ai (no install)

In Claude.ai, open Settings → Integrations → Add custom integration. Paste the MCP endpoint and your API key from the CloudClawer dashboard. Authorize. Done — Claude now shows a CloudClawer tool in every conversation.

endpointhttps://api.cloudclawer.com/mcp

authBearer <your-api-key>copy from dashboard → API Keys

Path B — CLI (terminal)

npm install -g cloudclawer
cloudclawer init

cloudclawer init writes the MCP config to your Claude Code settings automatically. No manual JSON editing.

Verify the connection by asking Claude: "Check my CloudClawer status."It should call mcp__CloudClawer__status and reply with your plan and memory count.

2

Set your daily schedule

A schedule tells CloudClawer whento fire Claude. Each slot holds one fire per day. You'll set a single slot to start.

Open the CloudClawer dashboard and go to Schedule
Click Add time slot
Pick a time — e.g. 09:00
Set the mode to queue
Save.

slot modes at a glance

dailyreusedsame task fires every day

queueconsumedtasks dequeue one at a time ← start here

emptyno injectionClaude fires with no pre-loaded task

Queue mode is the most useful starting point: tasks accumulate in a list and each run consumes the next one. When the queue empties, the slot idles until you add more.

3

Push your first task

Now give the routine something to do. You can do this two ways — from the dashboard, or by asking Claude directly.

Option A — dashboard

Go to Routines → Guided Queue, click Add task, and type your task description. It lands in the queue and will be injected at the next scheduled fire.

Option B — ask Claude

In any Claude conversation (with the MCP connected), say:

Add a task to my CloudClawer queue:
"Review open PRs and post a comment on any that are ready to merge."

Claude calls task_pushand confirms the task is queued. You'll see it appear in the dashboard immediately.

Good first tasks are concrete and self-contained: "Read the last 5 Linear issues and summarise what's blocking progress."Avoid tasks that need live user input mid-run — the agent executes alone.

4

Watch it fire

At the scheduled time, CloudClawer fires Claude automatically. Here's exactly what happens inside that call:

injectCloudClawer pulls your next queued task and injects it into the system prompt.

fireClaude receives the prompt and begins working — it can call all MCP tools (memory, task list, bash if on Plus).

briefWhen done, Claude posts a summary back. You see it in the fire log and in your dashboard.

dequeueThe task moves from queued → in_progress → done. The slot is now free for tomorrow.

You don't have to wait until tomorrow. On the fire URL page in your dashboard you can trigger a manual test fire any time — it runs your queue exactly as the schedule would, but right now.

Free accounts get 1 fire per day (resets at 00:00 UTC). Plus accounts get 15. The manual test fire counts against your daily limit.

5

Check the cost and fire log

Every run is tracked. Open Dashboard → Costs to see the session spend. You're looking for:

cost_usd$ amounttotal spend for this run

cache_pct0–100%≥80% is good — context is being reused

statusdone / errorfire log entry for the slot

The fire log (under Routines → Fire log) shows every run: timestamp, task that was injected, outcome, and cost. If a run errored, the reason is there — usually a tool permission issue or a task description that was too vague.

Low cache_pcton the first run is normal — the cache warms up as your schedule runs the same context repeatedly. By day 3 of the same slot you'll see cache rates above 80%.

What's next

You now have a working routine. From here there are two obvious directions:

Add a process gate

Attach a checklist to a task so the agent can't complete it without hitting every gate. Useful for deploys, releases, or any work with a must-not-skip step.

Climb to Warp 2

Wire CLAUDE.md, add your first role file, and give the agent a memory it carries between sessions. The Warp 2 guide walks each config item.

Connect your team org

Add teammates, share the schedule Gantt, and let the CoA attribute cost per ticket.