A PaaS that speaks MCP
Connect Claude Code, Cursor or any MCP client. Your AI builds, deploys, scales and fixes your apps on its own, with a database, logs and graphs included. You pay only for what runs, and every app gets its own *.appboox.com address.
Drag to look around, tap to deploy.
How it works
Request access, then tell your AI to connect with your API key, or add AppBoox as an MCP server in Claude Code, Cursor or any client that speaks MCP. One command, no SDK.
Tell your AI to put your app online. It talks the deployment through with you: checks the resources you have, suggests whether the app needs a database and creates it on the spot, how many VMs to split it into, and which subdomain you want, like pets-valley.appboox.com.
The app is up at https://name.appboox.com as soon as the build finishes, with Postgres, logs and graphs. Your AI keeps watching and fixes what breaks.
Built for AI agents
Deploys, builds, files, logs, metrics, databases, scaling and billing were designed as MCP actions from the start, with clear results and errors that say what to do next. Your AI operates the platform natively instead of clicking through a dashboard.
Each call your AI makes is logged: which tool, which app, how long, whether it failed. Watch it happen in the console.
Deploy, files, exec, logs, database and scaling are all tools. No server keys to hand out, no shell to babysit.
Give an agent a token for one app only. It can work on that app and cannot touch the rest of your account.
Before it creates anything, your AI asks you the right questions and gets a recommendation for the class, the size and the monthly cost. A few days later it checks the choice against real usage.
Everything included
Managed PostgreSQL 17 in seconds. Your AI creates, queries, resets and monitors its own databases (up to 2 per app), and the connection string is injected into the app for you.
Every app sits behind Cloudflare and our edge with automatic HTTPS and per-IP rate limits. Each app has its own firewall and SYN-flood guard, runs isolated with hard CPU, memory, disk and bandwidth limits, and cannot reach private networks.
Every app gets its own https://your-app.appboox.com address with HTTPS from the moment it is deployed. No public IP to rent, no certificates to renew, no load balancer to pay for.
Half the CPU price and well under half the memory price of Standard, billed on real use only. Ideal for prototypes, staging and test copies, demos, school and hobby projects, internal tools, bots and turn-based games: anything that is fine with a short cold start.
Measured on AppBoox's own cluster.
Live demo
This is Starfield Arena, a realtime multiplayer 3D game. An AI wrote it from a prompt and shipped it to AppBoox over MCP: a Node and WebSocket server, a three.js client, and high scores in managed Postgres. It is running live on the same platform that serves this page.
Pricing
Usage-based: you are billed on the CPU, memory and traffic your app really uses. Pick one of two classes per app. A sleeping app costs no CPU or RAM.
Memory is reserved for you and the app is never paused for capacity. For production sites, APIs, live game servers and workers that must not be interrupted.
RAM is billed at the larger of actual use and 25% of the limit.
May be paused when its node runs short of memory and wakes on the next request. Great for staging, batch jobs, low-traffic sites and turn-based backends.
Billed on actual use only.
Same on both classes: disk $0.10 per GiB-month, Postgres $0.50 per database-month plus $0.20 per GiB-month plus usage beyond the included allowance.
-
per month
-
-
per month
-
Estimates CPU and RAM only, before the free allowance. Disk and Postgres are billed separately.
Prices in USD; THB approximate (at 33 THB per USD). Every account gets a free allowance each month.
Observability
Logs, metrics, per-port traffic, an attack view and anomaly alerts are built in. These are real screenshots of the AppBoox console.
Safety
A separate token for each app, so a coding agent never needs your account key.
Nothing happens off the record. Each tool call and each shell command is kept and shown.
Sudden CPU, disk or outbound-traffic spikes are detected automatically and flagged to the platform operators, and a per-app attack view shows suspicious traffic.
Apps run finished builds, never compilers. Builds happen in a separate build service, so a runaway job cannot eat your app.
Request access, tell your AI to connect, and ask for your first app.
Read https://appboox.com/ai and connect AppBoox with this token: <your token>Works with Claude Code, Cursor, Codex, Gemini CLI, Copilot and any AI that can read a web page and add an MCP server.
Add AppBoox to your AI client yourself. Replace the key with the one from your AppBoox account.
claude mcp add --transport http appboox https://admin.appboox.com/mcp --header "Authorization: Bearer <your key>"Already have an account? Sign in
Any MCP client with Streamable HTTP and custom headers works.