For agent builders and vibecoders

Hosting your agent can operate

Claude Code, Codex, Cursor or a connector in claude.ai: your agent gets its own key with exactly the rights you choose, and every action lands in an audit log you can read.

Live in sixty seconds

  1. Put a site online for free and claim it in the portal.
  2. On the Agents tab, create a key with the rights your agent needs.
  3. Add the MCP server of your site to your tool:
claude mcp add --transport http dorelli https://mcp.dorelli.cloud/<slug> --header "Authorization: Bearer <key>"

The key is shown once. Connectors in claude.ai and ChatGPT do not need a key: they ask for consent through OAuth.

Rights, chosen by you

Six rights in version one. The key decides what a tool may do; a missing right returns an error with the page where you grant it.

  • deploy

    Publish a new version of the site.

  • status

    Read the address, status, plan and expiry.

  • logs:read

    Read what happened to the site: deploys, moderation, payments, tickets.

  • forms:read

    Read form submission counts per day. The contents go to the owner by email.

  • audit:read

    Read the audit log of the site.

  • limits:read

    Read the limits of the plan and today's usage.

Coming later, after proof with real agents: rollback, database, secrets, restart, and building blocks within a budget on Platform.

Limits per plan

These numbers come from the same file the code enforces them with. What is not in this table is not a promise.

PlanKeys per siteWebhooksAudit log keptDeploys per day
Website3130 days50
App5390 days20
Platform101090 days + CSVunlimited within the cap

Audit log

Who did what, with which key, with the outcome and the time. Also the refused actions. Never the contents of a deploy or a form submission.

Webhooks

A POST to your URL when something happens on your site, signed with HMAC-SHA256. Three attempts, then we give up and say so.

deploy.klaar · moderatie.geblokkeerd · site.verloopt · plafond.80

OAuth for connectors

claude.ai and ChatGPT connect through a consent screen: you pick the site and the rights. The access token is an agent key that expires after a day and is refreshed by the connector.

AGENTS.md per site

One file for your repository that tells any coding agent how to deploy here, where the logs are and which limits apply. Download it from the Agents tab.

Your agent, your responsibility

An agent with a key acts on your behalf. We log every action and the limits are hard, but what it does within them is up to you. Keep the key secret and give it no more rights than it needs.

The terms article on agent keys