Cloud for Bots docs

One key and one MCP server give your bot sandboxes, a logged-in browser, storage, secrets, queues, cron, locks and a human when it gets stuck. Bots get this guide automatically when they connect. Plain text: /bots.md

Quickstart

1. Get a key. Enter your email on the homepage. Your key shows once and is copied for you. No card.
2. Connect your bot. In Grok Bot (or any MCP client) add the server:

   URL:     https://cloudforbots.com/mcp
   Header:  Authorization: Bearer <your key>

3. First win. Ask your bot: “Run python in my Cloud for Bots sandbox and add the numbers 1 to 100.” It should answer 5050. Or from a terminal:

   curl -s https://cloudforbots.com/api/v1/sandbox \
     -H "Authorization: Bearer $CLOUDFORBOTS_API_KEY" \
     -d '{"action":"run","cmd":"python3 -c \"print(sum(range(101)))\""}'

Want to try first? The homepage has a live sandbox you can run without a key.

Connect your bot

Step-by-step with copy buttons: /start.

Which tool when

• sandbox: heavy or long work (code, installs, data jobs). It pauses instead of dying, so files survive. Each run is up to 120 s; use background: true for longer jobs and check back.
• store: save progress and results (set/get). Save after every step so you can resume after a crash or a new session.
• inbox, queue, cron: wake up later. inbox gives you a URL that webhooks and other bots can POST to; messages land in queue inbox.<name>. cron runs on a schedule or a one-off timer. Read with queue pull, then ack when done.
• lock, rate: stop bots colliding. lock = one writer at a time. rate = shared limits like "6/6h" across all your bots.
• secrets + fetch: never paste API keys into chat or code. Store once with secrets set (with allowed_hosts), then call APIs with fetch using {{secret:NAME}}.
• browser: sites that need a login. Sessions keep cookies between calls.
• handoff: when a human is needed (captcha, 2FA). Watch-only on free keys.
• ledger, replay, loop_*, budget_*: safety. Record runs in ledger, check repeats with loop_check_step, cap spend with budget_open_budget and budget_reserve, review with replay.

Pattern 1: long job with resume

1. store get ns jobs, key <job> -> last finished step (none = start at 0).
2. For each remaining step: sandbox run it, then store set {"step": n, "result": ...}.
3. If you crash or the chat ends, start again at step 1. You skip finished steps.

Pattern 2: webhook-driven bot

1. inbox create name orders -> give the URL to the other service.
2. cron set (or timer) with {queue: "wake.orders"} if you need a nudge, or just check on each run.
3. queue pull inbox.orders, handle each message, then ack its id. Un-acked messages come back.

Pattern 3: safe posting with a rate limit

1. lock acquire post (ttl_s 60). If you do not get it, another bot is posting: stop.
2. rate take posts limit 6/6h. If not allowed, wait retry_after_s.
3. fetch the posting API with {{secret:POST_TOKEN}}.
4. ledger record the run, then lock release with the token.

Rules of thumb

• Save to store before and after anything slow.
• Ack queue messages only after the work is done.
• Ask your human before spending money or posting publicly.
• Full reference: https://cloudforbots.com/llms-full.txt

Custom instructions for your bot

Paste this once so your bot knows when to reach for each tool.

• Grok Bot: paste it into a chat, or wherever you keep its standing instructions.
• Claude: Claude app: project instructions. Claude Code: add it to CLAUDE.md.
• ChatGPT: Settings → Personalization → Custom instructions.

You have Cloud for Bots tools (MCP server "cloud-for-bots"). Use sandbox for heavy or long jobs. Save progress to store after each step and resume from it. Use inbox, queue and cron to wake up later. Take a lock and a rate slot before posting or other shared actions. Use secrets + fetch instead of pasting API keys. Use browser for logged-in sites and handoff when a human is needed. Ask me before spending money or posting publicly. Full guide: https://cloudforbots.com/bots.md

Tool reference

Every tool works three ways: as an MCP tool, as REST, and from the TypeScript SDK. All calls use Authorization: Bearer <key>.

curl -s https://cloudforbots.com/api/v1/sandbox \
  -H "Authorization: Bearer $CLOUDFORBOTS_API_KEY" \
  -d '{"action":"run","cmd":"python3 -c 'print(5050)'"}'

curl -s https://cloudforbots.com/api/v1/browser \
  -H "Authorization: Bearer $CLOUDFORBOTS_API_KEY" \
  -d '{"action":"open","url":"https://example.com"}'

curl -s https://cloudforbots.com/api/v1/store \
  -H "Authorization: Bearer $CLOUDFORBOTS_API_KEY" \
  -d '{"action":"set","ns":"jobs","key":"daily","value":{"step":3}}'

curl -s https://cloudforbots.com/api/v1/secrets \
  -H "Authorization: Bearer $CLOUDFORBOTS_API_KEY" \
  -d '{"url":"https://api.example.com/me","headers":{"Authorization":"Bearer {{secret:EX_TOKEN}}"}}'

curl -s https://cloudforbots.com/api/v1/queue \
  -H "Authorization: Bearer $CLOUDFORBOTS_API_KEY" \
  -d '{"action":"pull","queue":"inbox.orders","max":5}'

curl -s https://cloudforbots.com/api/v1/cron \
  -H "Authorization: Bearer $CLOUDFORBOTS_API_KEY" \
  -d '{"action":"set","name":"daily","schedule":"@daily","request":{"queue":"wake.daily","message":"go"}}'

curl -s https://cloudforbots.com/api/v1/lock \
  -H "Authorization: Bearer $CLOUDFORBOTS_API_KEY" \
  -d '{"action":"take","name":"posts","limit":"6/6h"}'

curl -s https://cloudforbots.com/api/v1/handoff \
  -H "Authorization: Bearer $CLOUDFORBOTS_API_KEY" \
  -d '{"action":"create","kind":"2fa","note":"Bank login code"}'

curl -s https://cloudforbots.com/api/v1/vault \
  -H "Authorization: Bearer $CLOUDFORBOTS_API_KEY" \
  -d '{"action":"login","name":"crm","url":"https://crm.example.com/login","form":{"email":"{{username}}","pw":"{{password}}"}}'

curl -s https://cloudforbots.com/api/v1/ai \
  -H "Authorization: Bearer $CLOUDFORBOTS_API_KEY" \
  -d '{"model":"grok-4","messages":[{"role":"user","content":"hi"}]}'

curl -s https://cloudforbots.com/api/v1/ledger \
  -H "Authorization: Bearer $CLOUDFORBOTS_API_KEY" \
  -d '{"action":"rollup","days":7}'