AI-native, because the platform was designed for agents from the start
Most hosting bolts an assistant onto a dashboard. We expose the platform itself over an open protocol, so your AI tool of choice drives it directly.
The Model Context Protocol is the open standard for connecting AI agents to systems. Our MCP server is a thin protocol adapter sitting over the same engine API the dashboard uses — the same action catalog, the same permissions, the same audit trail. That means there is no second, weaker set of AI-only logic to get out of step, and no per-tool integration to maintain. We support the protocol; whatever agent you use next works too.
The three primitives are all there. Tools are the actions: clone to staging, push to live, purge cache, restore from Wayback, run wp-cli, connect a git repo and deploy or roll back, edit DNS, issue certificates, take backups, read SEO data, check billing. Resources are read-only context — site health, config, recent logs, metrics, KB articles — so the agent diagnoses with real data before it touches anything. Prompts are reusable workflows we publish, such as diagnosing a site or preparing a migration.
Safety is the reason it can be allowed near live infrastructure. Authorisation is OAuth 2.1, tokens are scoped to your organisation and your role permissions with row-level security enforced in the database, and every token is per-tool and revocable. Destructive actions require explicit confirmation, spend caps bound anything that costs money, retries are idempotent, and every action is written to the audit log with the identity, the tool, the arguments and the result.
- Connect from the dashboard: generate an endpoint, then paste a one-click config snippet for Claude Code, Cursor, ChatGPT or Claude Desktop.
- Interactive by design — AI actions stream and are never queued behind a background job system.
- Separate sandbox and production tokens, so an agent can be tried without pointing it at live sites.