Restore on a new server

Your jobs live in your repos, so a new server can rebuild them. restore recreates the jobs a repo’s .claude/jobs.json lists that Paperclip doesn’t have.

Steps

1. Deploy the template again.

2. On the new server’s /setup page, connect Claude and GitHub, and add the same repos.

3. Open a chat in CloudCLI. Claude sees that your repos record jobs this server doesn’t have, and offers to restore them. Or ask: “restore my jobs”.

4. Claude shows what it would recreate first:

   acme/website: would recreate "Weekday open issues digest" (0 8 * * 1-5 America/Toronto, active); "Daily uptime check" (0 7 * * * America/Toronto, paused, needs STATUS_WEBHOOK_URL). Run again with --yes to do it.

5. Once you agree, Claude runs restore --yes:

   acme/website: recreated "Weekday open issues digest" (0 8 * * 1-5 America/Toronto, active); "Daily uptime check" (0 7 * * * America/Toronto, paused).
   "Daily uptime check" needs STATUS_WEBHOOK_URL, which Runner doesn't have yet. The user adds it on Runner's Harness / Runtime page in Paperclip (https://your-paperclip.up.railway.app/JOB/agents/<agent id>/runtime): under Environment variables, Add variable, type the name, then choose "Organization secret" and "Create new secret..." for a key or token, or "Text value" for a plain setting such as a chat ID, and Save changes.
   Recorded in acme/website's .claude/jobs.json: commit 5e6f7a8, pushed.

6. Add the secrets the list names, on the Runner’s Harness / Runtime page in Paperclip, under Environment variables. See Secrets.

7. Test a job or two: “run the digest now and show me the log”.

What restore keeps

• Each job’s title, prompt, schedules, time zones and status: paused jobs come back paused.
• The agent: a job recorded for Codex goes to the Codex agent. Connect Codex on /setup first, or restore skips it and says so.
• The secret names, so the script tells you which values to add.

Jobs get new IDs in the new Paperclip. The record is updated with them and pushed.

What restore doesn’t do

• It never runs on its own. Without --yes it only shows what it would do, and Claude asks you first.
• It skips jobs archived on this server’s Paperclip. Run sync to drop them from the record.
• It doesn’t bring back run history. Past runs stay in the old server’s Paperclip.
• It doesn’t copy secret values. You add them again on the new server.

One server per repo

Two servers with the same repo and the same jobs run each job twice. Once the new server runs your jobs, remove the old project in Railway. Don’t pause or archive the jobs on the old server through Claude: those changes are written to the record in your repo, which the new server uses too.

See also Move to a new server.