Run your infrastructure from an AI client or a terminal
The same account, the same permissions, the same limits. Connect an AI assistant over the Model Context Protocol, or install a single binary and use it like any other command line tool.
Connect an AI client
Point Claude, ChatGPT or any Model Context Protocol client at our hosted endpoint. You approve the connection in your browser and choose what it may do.
https://mcp.techbeaver.io/mcpInstall the CLI
One static binary. No runtime to install, nothing to configure before it works.
curl -fsSL https://techbeaver.io/install.sh | shConnecting an AI client
Nothing to install. Your client registers itself and sends you here to approve it.
- 1
Add the connector in your AI client
Wherever your client lists connectors or MCP servers, add https://mcp.techbeaver.io/mcp. - 2
Approve it in your browser
Your client opens a TechBeaver page listing exactly what it is asking for. Untick anything you would rather it could not do. If you are already signed in, this is one click. - 3
Ask it to do something
Try “list my apps” or “deploy my API and tail the build log”. It sees only what you granted.
Using the CLI
Sign in
beaver auth login
beaver auth whoamiOn a server with no browser, beaver auth login --no-browser prints a short code to enter on another device.
Do the everyday things
beaver projects list
beaver apps deploy --wait
beaver apps logs --follow
beaver db create my-database
beaver billing invoices listPin a project to a directory
Drop a beaver.toml beside your code and the commands get shorter.
project = "my-project"
app = "my-api"Script against it
Output is JSON automatically when it is not going to a terminal, so a pipe never receives a table you have to parse.
beaver apps describe my-api | jq .data.url
beaver apps list --format value(name,status)Deploying from CI, with no token to store
A pipeline does not need a secret from us. GitHub Actions can prove which repository and branch a job is running for, and we accept that proof directly. Nothing goes into a repository secret, nothing expires every 90 days, and nothing useful leaks if a build log is public.
- 1
Name the repository in your console
Under AI & API access, add the repository, and optionally the one branch or environment allowed to deploy. Wildcards are not accepted: you name one repository. - 2
Add one line to your workflow
id-token: write is what lets the job ask GitHub for its identity. Without it there is nothing to prove. - 3
Run beaver
No login step. The CLI notices it is on a runner, presents the job’s identity and gets a credential that lasts an hour.
.github/workflows/deploy.yml
jobs:
deploy:
runs-on: ubuntu-latest
permissions:
id-token: write
contents: read
steps:
- uses: actions/checkout@v4
- run: curl -fsSL https://techbeaver.io/install.sh | sh
- run: beaver apps deploy --waitA deploy made this way is attributable to the workflow run that made it, which a shared token never could be. And a pipeline still cannot delete an app or a database without you approving it in a browser.
What an AI agent can never do on your account
These are enforced by our servers, not by the client. An out-of-date, modified or home-built client gains nothing.
It cannot delete anything on its own
Every deletion creates a request you approve in your browser, naming the exact thing being deleted. An agent cannot approve its own.
It cannot complete a payment
It can prepare one and hand you the link. Only you can pay it, on the payment page.
It cannot spend your account credit
Moving money without a payment page is a permission nothing can currently be granted.
It cannot reach anything administrative
There are no administrative permissions in the vocabulary at all, so no combination of what you grant adds up to one.
It only ever sees what you ticked
A permission you did not grant is not merely refused. The tool does not appear, so the agent cannot suggest it.
Prefer to keep the AI on your own machine?
The CLI is also an MCP server. Point a local AI client at it and it uses the credential already on your machine, with no hosted connection at all.
In your AI client's configuration
{
"mcpServers": {
"beaver": { "command": "beaver", "args": ["mcp"] }
}
}Manage your connections
Every connected client, every token and every pending approval is listed in your console, and you can revoke any of them at any time.
