The agent-first cloud · early access

The cloud your agents operate.

PromptShip is an agent-first deployment platform. Agents build, verify against real data, promote to production, and roll back, all from Claude Code, Cursor, and Codex through one MCP server on PromptShip's managed runtime.

Free during early access.

Already have access? Install the MCP:

$ claude mcp add --transport http promptship https://mcp.promptship.dev/mcp

From the team behind Pyde · Standard Docker, no lock-in

promptship · agent pipelinelive

$ list_agent_actions --agent claude

claude/tok_a91fcreate_envpr-42ok12:03:58

claude/tok_a91fclone_dbprod → pr-42ok12:04:11

claude/tok_a91fdeploypr-42ok12:04:49

claude/tok_a91frun_checkspr-4212 passed12:05:37

claude/tok_a91fpromoteprodok12:06:02

claude/tok_a91fverify_prodprodok12:06:20

Agent-native infrastructure

The agent-managed cloud.

Agents don't just deploy here. They operate the whole environment: spin up a full stack, clone it with real data, verify, promote to production, roll back. Every action attributed to the agent that took it.

Replacing a stitched E2B + Neon + Fly stack? See the comparison →

Isolated full-stack environments

App, Postgres, and cache in one unit your agents operate. Every environment carries a TTL with automatic cleanup and scoped credentials.

Clone-with-data

Rolling out in open beta

Clone a running environment including its database in one call. Agents iterate against real data, not fixtures.

Promote with rollback

Rolling out in open beta

Atomic promotion of a verified environment to production. Instant rollback when it is wrong.

Agent identity + action ledger

Rolling out in open beta

Every action bears the agent’s own token. “Show me everything this agent did” is a one-call answer.

A public URL for every app

Every app your agent builds gets its own URL, and it outlives the session. Share it, keep iterating on it.

Sleeps at $0

Rolling out in open beta

Idle apps cost nothing. The URL stays live and the app wakes on the next request.

One conversation, full stack

Deploy the app, attach Postgres, Redis-compatible cache, or ClickHouse, and get SSL, all through the MCP server without leaving your session.

For Claude Code & Codex power users

Your Claude Code session's apps, persistent and shareable.

The apps you build in a session shouldn't die with it. PromptShip turns them into real deployments: a live URL, a database, and a runtime that keeps running after you close the terminal.

The real thing

Watch a real session

A real Claude Code session deploying through the PromptShip MCP.

Claude Code · PromptShip MCP

The pipeline

Push your code. We build, ship, and scale it.

From AI-generated code to a live URL, through one MCP server.

Your code
Build
Runtime
Production app
Preview env with real dataOpen beta
Instant rollbackOpen beta

PromptShip operates the whole pipeline end to end: build, runtime, and managed databases. The output is a standard Docker image. If you ever leave, it runs anywhere Docker runs.

The platform

One platform for everything an AI IDE ships

Backend services, frontends, databases, deployed the same way.

Backend and frontend, one platform

Long-running services, Next.js SSR, static sites, all through one MCP server. Same deploy flow, same dashboard, same bill.

Persistent containers

No cold starts. WebSockets, background workers, cron, they all just run. Real always-on processes, not request-scoped functions.

Postgres, Redis, ClickHouse, provisioned via MCP

One command, attached to your app. Managed Postgres, a Redis-compatible cache, and ClickHouse analytics, no third-party setup, no separate bills.

Transparent tier pricing

No credits, no tokens, no per-request fees. Pick a tier, see the price. You set the autoscaling limits; you control the bill.

Environments that clean up after themselves

Every environment carries a TTL and is garbage-collected automatically. Scoped credentials, nothing left behind.

Custom domains with auto-SSL

Point your domain, add a CNAME, done. Certificates renew automatically, no DNS rituals, no expired-cert pages.

Compare

Outgrowing Vercel?

Move the backend first. Bring the frontend over when you're ready.

PromptShipVercel
Persistent containers (no cold starts)
WebSockets
Persistent background workers
Managed Postgres attached
Managed Redis-compatible cache
Managed ClickHouse analytics
Predictable tier pricing (no per-request fees)
No platform lock-in (standard Docker image)

Vercel is excellent for frontends and edge functions. For long-running backends (WebSockets, workers, attached databases) PromptShip is the deploy target. Migration guide →

Evaluating a stitched E2B + Neon + Fly stack? See the comparison →

Pricing that fits in a sentence

Tiers start at $10/mo for backend + Postgres. Free during early access. No credits, no tokens, no per-request fees.

See all tiers →

FAQ

Frequently asked questions

Wait, is this a prompt management tool?

No. PromptShip is a deployment platform. Tools like PromptLayer manage LLM prompts; PromptShip runs the apps you build with AI IDEs, backends, databases, custom domains.

Is PromptShip a wrapper around Cloud Run or another PaaS?

No. PromptShip operates its own platform, build pipeline, runtime, and managed databases. We're not a layer above Cloud Run, Vercel, or Render. Like every PaaS, the platform itself runs on managed cloud infrastructure underneath (DigitalOcean today), but that's IaaS, the same way Vercel runs on AWS. If you ever leave, your Docker image runs anywhere Docker runs.

How is this different from Replit or Bolt?

Replit and Bolt are AI app builders that include hosting. PromptShip is the infrastructure AI IDEs deploy to. Different category, different comparison set: Cloud Run, Fly, Railway, Vercel, not Replit, Bolt, v0.

Can I migrate from Vercel gradually?

Yes. Most teams move the backend first since that's where Vercel struggles (WebSockets, background workers, attached Postgres/Redis/ClickHouse, jobs longer than a request). Once that's stable you can bring the frontend over too, since PromptShip runs Next.js SSR and static sites the same way it runs the backend. Step-by-step at /migrate/vercel.

How do I deploy from Claude Code, Cursor, or Codex?

Add the PromptShip MCP server to your AI IDE with one command, connect your GitHub repo, and deploy, all from a single conversation. PromptShip builds a Docker image, provisions your database, sets up SSL, and gives you a live URL.

Does PromptShip include a database?

Yes. Every app can attach managed PostgreSQL, a Redis-compatible cache, and ClickHouse for analytics. More databases on the roadmap. Provisioned via MCP, no third-party setup, no separate billing.

What about WebSockets and background jobs?

PromptShip runs persistent containers, not request-scoped functions. WebSockets, Server-Sent Events, background workers, and cron jobs all work with no special configuration and no timeouts.

What does 'sleeps at $0' mean?

Idle apps scale to zero and cost nothing while asleep. The public URL stays live, and the app wakes automatically on the next request. Rolling out in open beta.

Can agents test against real data?

Yes. Clone a running environment together with its database in one call. Agents build and verify against real data in an isolated copy that carries a TTL and is cleaned up automatically. Rolling out in open beta.

How do promote and rollback work?

Promotion atomically makes a verified environment the production one; if something is wrong, rollback restores the previous version instantly. Rolling out in open beta.

How do I see what an agent did?

Every action on PromptShip is taken by an identified agent bearing its own token. 'Show me everything this agent did' is a one-call answer via the action ledger, exposed over MCP and in the console. Rolling out in open beta.

How much does PromptShip cost?

Creating a team is free. Infrastructure is billed per resource. A backend + Postgres tier starts around $10/month once pricing kicks in. No credits, tokens, or per-request fees. Early access is free; no payment method required.

Why is there a waitlist?

We're scaling onboarding gradually. Most requests get access within a few days. If you've already added the MCP to your IDE, your account is automatically queued, no separate signup needed.

Early access

Now in early access

Free during early access. Open beta is rolling out; waitlist members get in first.