The agent-first cloud · early access
The cloud your agents operate.
Full-stack apps your agent can deploy, debug and ship, with you approving production.
claude mcp add --transport http promptship https://mcp.promptship.dev/mcpYour agent reads prod.
It can't break it.
Ask a product question. Your agent queries prod data, reads the logs and finds the cause.
- Reads run without asking you
- Postgres, Valkey and ClickHouse
- Writes to prod wait for you
Production waits
for your yes.
Dev and previews are your agent's. Prod waits for you.
- Ships the exact commit you approved
- Scope an agent to previews and prod is off limits
Three agents,
three ideas.
Clone prod, code and data, into isolated environments. Each agent tries its idea on real data. You keep the winner.
- Code, data and config, copied
- Nothing in a clone reaches your users
- Clones expire on their own
Every agent
signs its work.
Each agent connects with its own name. Every call lands in the ledger with its task and who approved it.
- One named identity per agent
- Every action grouped by task
- Every approval records who said yes
Ledger · geo-auditor · today
claude-code
first-audit
codex
cache-stats
Questions
How is PromptShip different from Railway, Vercel or Fly?
PromptShip is built for agents running your app. Each agent gets its own scoped identity, production changes wait for your approval, and every action is recorded under its name.
Which AI tools does PromptShip work with?
Claude Code, Cursor, Codex, and any client that speaks MCP. Adding the server is one command. There is also a CLI.
What can I run on PromptShip?
Web services, workers, cron jobs and WebSockets on persistent containers, built from your Dockerfile or detected for Python, Node.js, Go, Bun and static sites. Each environment can attach Postgres, Valkey or ClickHouse.
How do approvals work?
Anything your agent does to production waits for your approval: deploys, rollbacks, secrets, domains and write queries. You approve in the dashboard, and approval ships the exact commit you reviewed. Reads never need approval.
How does pricing work?
You pay for the resources you run, prorated to five minutes, plus a plan. Free includes $25 of usage a month, Hobby is $5 a month with $5 of usage included, and Pro is $20 a month with usage billed on top. No per-request fees. Full rates at /pricing.
What is available now, and what is open beta?
Deploys, databases, domains, agent tokens, approvals, the ledger, and rollback are live. Clone-with-data and tested-build promotion are open beta.
Now in early access
Free during early access. Waitlist members get in first.