Enforced runbooks for AI agents

The checklist
your AI
can't skip

Write the procedure once. Your AI runs it in order and stops at the steps only you can approve.

Release 1.27.3 pre-flightRUN #47

run in progress: order enforced

ORDERENFORCED

A step cannot complete before the steps it depends on. Skipping is refused by the server, not discouraged by a prompt.

EVIDENCEREQUIRED

Steps you mark as needing proof cannot close empty-handed. Command output, links, and hashes stay attached to the run forever.

APPROVALHUMAN-ONLY

Gated steps stop the run until you approve in your browser. Agent tokens cannot wave themselves through; the database refuses.

Airborne in three steps

1
Mint an agent token
Sign in, open Settings → Agent tokens, name one, copy it. Revoke it any time; the agent is locked out instantly.
2
Connect your AI
Checklists.io speaks MCP, the standard way assistants use tools. One command wires up Claude Code; plain HTTP works for scripts and CI.
```
claude mcp add --scope user --transport http checklists \
  https://checklists.io/api/mcp \
  --header "Authorization: Bearer ck_…"
```
3
Run with the keys in your pocket
Describe your procedure and let your AI create the runbook. Watch runs live, read the evidence, and approve the dangerous steps from any browser.

Built for procedures with consequences

Software releasesDatabase migrationsProduction deploysIncident proceduresInfrastructure changesAnything you double-check