What you need
- A Techbeaver account. If you do not have one, your AI client can send you to sign up the first time it connects.
- Your code in a Git repository. For a private GitHub repository, install the Techbeaver GitHub App on it first; the agent checks access before it creates anything, so a missing permission is reported up front rather than as a failed build.
- An AI client that supports MCP (the Model Context Protocol), such as Claude or ChatGPT, or the
beaverCLI.
Step 1: Connect your AI client
In your client’s connector or MCP server settings, add https://mcp.techbeaver.io/mcp. Your browser opens a Techbeaver page listing exactly what the client is asking to do. Untick anything you would rather it could not, then approve.
Step 2: Ask it to deploy
Describe what you want in plain words. For example:
Try asking
Behind that one request, the agent works through the same steps you would:
- checks it can read the repository and lists the plans with their current prices;
- creates the project and the app, and for a monorepo points it at the right folder;
- hands you a payment link for a priced plan, because an agent can never pay on your behalf;
- follows the build and tells you the live URL once it is up.
Step 3: Let it fix what breaks
When a build fails, the agent reads the same build log you would, and can change the settings that most often cause it: environment variables, the start command, the port the app listens on and the root directory. Then it redeploys and checks again. It can also roll back to a previous release, read runtime logs and metrics, and add a custom domain with the DNS records to set.
Secrets in a conversation
What stays with you
- Payments. The agent prepares a payment and gives you the link. Only you can complete it, and it cannot spend your account credit.
- Deletions. Deleting an app, a project or a database creates a request you approve in your browser, naming exactly what will be removed. Stopping an app is reversible and needs no approval.
- Access. Every connected client is listed under AI & API access in your console, and you can revoke it at any time.
Prefer the terminal?
The same deploy is one command with the beaver CLI: beaver apps deploy --wait. From GitHub Actions it needs no stored token at all.
