Agent-first hosting · run in the EU
Hosting your coding agent can operate.
Your agent builds the app. agentbuild is where it goes live: the agent connects over MCP, publishes your GitHub repository and hands you an HTTPS address – with Postgres, deploys on every push and nothing to click through.
Start free – 2 apps Read the quickstart →
Sign in with GitHub. No credit card.
you Publish github.com/alice/shop on agentbuild. agent create_app(app="shop", source_repo="https://github.com/alice/shop") → building from source … app_status(app="shop") → ready: true agent It's live: https://shop-alice.build.agenthotel.org Postgres is in DATABASE_URL. Every push to main deploys a new version.
From repository to live app in three steps
Create an account
Sign in with GitHub. You get an API key for your agent.
Connect your agent
Add agentbuild as an MCP server – one command or a few lines of config in Claude Code, Codex, Cursor, opencode, Hermes and others.
Ask it to publish
“Publish this repository.” The platform builds it, runs it and returns the address. After that, every push deploys.
What every app gets
An HTTPS address
<app>-<you>.build.agenthotel.org from the first deploy, with a valid certificate.
Its own Postgres
A real PostgreSQL database per app, handed to the app as DATABASE_URL. Optional.
Deploy on every push
Built from your GitHub repository – your Dockerfile, or detected automatically for Node, Python, Go and static sites.
Previews per pull request
Each open pull request can get its own address and its own database.
Scale to zero
Idle apps sleep and wake on the next request, so side projects cost nothing to keep around.
Your own domain
Two DNS records and a verification. The certificate is issued for you.
Built for agents, not for dashboards
Cloud platforms were designed for people clicking through consoles. An agent needs something else: a small, predictable set of tools, clear answers and hints when something is wrong.
agentbuild's whole surface is about twenty MCP tools – create_app,
app_status, app_logs, add_domain … – and every agent gets
the same ones. Switch agent tomorrow; your apps do not notice.
- Any agent. Anything that speaks MCP over HTTP.
- Or no agent. The portal can do everything the API can.
- No lock-in. Standard containers, standard Postgres, your code in your GitHub.
- Run in the EU. Servers in Finland and Germany, on open-source infrastructure.
- Isolated. Each account runs in its own namespace with its own limits and network rules.
Pricing
Free
€0
For trying it out and for small projects
- 2 apps
- Postgres per app
- HTTPS and custom domains
- Deploy on every push
- Preview per pull request
- Scale to zero when idle
- No credit card
Pro
Coming soon
For products with real users
- 20 apps
- Always-on apps
- More CPU, memory and storage
- Everything in Free
Questions and answers
What is agentbuild?
agentbuild is a hosting platform that coding agents operate directly. Your agent connects over MCP with an API key, publishes a GitHub repository as an app, and gets back a live HTTPS address. You manage the account in a portal; the agent does the building and deploying.
Which agents work with agentbuild?
Any agent that can use a remote MCP server: Claude Code, Codex CLI, Cursor, VS Code with Copilot, opencode, Hermes Agent, Windsurf, Claude Desktop, OpenClaw and others. See Connect an agent for copy-paste configuration.
What does every app get?
An HTTPS address, an optional Postgres database of its own, automatic builds and deploys on every push to GitHub, a preview address per pull request if you want one, scale to zero when idle, logs, and the option of your own domain.
How is this different from Vercel, Supabase or Lovable?
Those are built around a human in a dashboard. agentbuild is built around an API that an agent calls – the same few tools whichever agent you use – and it runs ordinary containers with ordinary Postgres on servers in the EU, so nothing about your app is tied to the platform.
Do I have to use an agent?
No. Everything an agent can do, you can do in the portal: create apps from a repository, add domains, read logs, manage API keys. Both ways work side by side.
Where is my app and data hosted?
On servers in the European Union (Finland and Germany, at Hetzner). The platform is built on open source: Kubernetes (k3s), Flux, CloudNativePG, KEDA and Traefik.
What kind of apps can I run?
Anything that runs as a container and listens on a port: Node, Next.js, Vite and other static sites, Python, Go and more. A Dockerfile is used if the repository has one; otherwise the platform works out how to build it.
What does it cost?
Two apps are free, with no credit card. A paid plan with more apps and always-on capacity is being prepared. See Pricing.
Can I use my own domain?
Yes, on the free plan too. Add the domain, create two DNS records (a TXT record that proves the domain is yours and a CNAME or A record), verify, and the certificate is issued automatically. See Domains.
Is there lock-in?
No. Your code stays in your GitHub repository, your app is a standard container and your database is standard Postgres. You can run the same thing anywhere.