What's new: your own AI agent and app secrets
A lot changes in Basable every week, and most of it is invisible: a fix in the deploy pipeline, a capacity calculation that got more honest. This post is the part you can see. Everything below is live on basable.com today.
Build with the AI agent you already use
Basable now has an MCP server. If you already work in Claude Code, Claude in the browser or Cursor, you can connect it to Basable and let it create projects, write their code, build it and deploy it, the same way Basable’s own agent does in the editor.
Your agent does the thinking; Basable builds, deploys and hosts. It reads the
same rules our own agent follows (the CI workflow, the Kubernetes manifests,
the security settings), so what it makes deploys the same way. An agent in a
terminal works in a clone and pushes to dev itself. An agent in the browser,
which cannot run code, sends its changes to Basable to build, and nothing
reaches dev without a green build.
Signing in is OAuth, so there is no key to copy. In Claude Code it is one command:
claude mcp add --transport http basable https://basable.com/api/mcp
Every app you connect is listed in your account settings, where you can disconnect it. The MCP guide has the setup for each client and the full list of tools.
Start from the home page
You can start before you have an account: the home page now opens on a prompt box. Type what you want to build, and the idea follows you through sign-up into the editor’s chat.
Secrets your app needs, and nobody else sees
Real apps need values only you have: a Stripe key, an SMTP password. The assistant now writes the app to read them from a secret, and tells you in the chat which keys it needs and where to get them. The deployment then waits at a new step with an Add secrets button until every key is set.
The values live in one place: your environment’s own cluster. Basable does not keep a copy in a database, a log, a CI artifact or anywhere else, and they never pass through the chat. On the live environment you can copy them over from preview.
Builds the agent checks before they land
When the assistant’s change builds green, it no longer merges by itself. The
agent first sees every file the build added (a compiled binary that slipped
into the commit, say), removes what does not belong, and only then merges it
into dev. While that build runs, the session says so, and you can cancel it.
Smaller things
- Redeploy. The deployment drawer has a button that runs the newest deployment of an environment again, without a new commit. Useful when the last one failed for a reason that has since gone away.
- Merge & Deploy waits for preview. The button opens once your code has been deployed to preview at least once, so nothing goes live that never ran.
- Bring code that already exists. Connect a repository that already holds an app, and the editor offers to get it running on preview: the assistant adds whatever is missing to build and deploy it.
- The preview shows inside the editor. Generated apps now allow the editor to show them in its preview pane, instead of the browser refusing to.
- Self-repair. You get full admin access to your cluster, which includes deleting the parts that make it work, like its DNS. A deployment now notices and puts them back before it continues.
Tell us
Plan mode, for larger projects in Rust, also moved a long way this week. It gets a post of its own. If something here does not work the way this post says, tell us: that is exactly what the alpha is for.