Settings › Advanced is the developer surface. The tabs need the develop capability (or a manager role); Hooks, Tokens, Identities and Secrets additionally need a paid plan, and the SDK tab's content needs an Enterprise or Partner space (the space tier) — in any other space the tab says whom to contact.
Hooks#
The project's outbound subscriptions — the hook feed. Add hook leads with the standard multi-event form (name, endpoint, signing key, an events multi-select; the server group is the render-server feed) and keeps the per-event legacy form behind it. Each hook has a Verify button that sends one signed probe to the endpoint and reports what came back.
Each row also shows the hook's last delivery status once it has fired: ok, failing (the last delivery failed) or down (three or more failures in a row), with the failure's status code or reason in place of the endpoint. The hook's page has the detail — last attempt, last success, consecutive failures, the error line. Nothing is retried on your behalf: fix the endpoint, Verify it, and the status clears on the next real delivery (a verify alone does not reset it). Disabled hooks show no status.
Tokens#
Project access tokens: mint one with a note, a role and an access tree (which route groups it may call, with per-level fallbacks), reveal a stored token later, edit its note, revoke it. The token marked Website is the platform-managed serving token behind web hosting — revoke it only through the hosting rotation. This is the gate the docs site's Try-it rail reuses for its token picker. See Authentication.
Identities#
Customer-scoped identity tokens minted from a parent token, with their origin allowlist; revoking the parent revokes them all.
Secrets#
Encrypted key–value secrets the platform's own features read (payment keys, SMTP, analytics credentials) — write-only; a stored value is never displayed again.
SDK#
The developer hand-out: how to set up the Starter, with the environment variable your minted token goes into. A manager on a paid plan also sees the project's MCP token for the active environment (the access token noted MCP, public + admin access plus the console grant that lets an agent set up the pinned menu and the dashboard) — masked, with Reveal and Copy, or a Generate MCP token button while none exists; a token generated before 2026-10-15 lacks the console grant, and Add the console grant adds it in place (the token value stays the same) — and Download ready package: the starter with the project id and that token already written into .env, so the only steps left are unpacking and opening the folder with an agent (the MCP server inside needs no npm install). The package contains the token; treat the zip as a secret. To connect a tool as yourself rather than with the project's token, use a personal access token from your Profile.
MCP#
Connecting an AI client to the project through the hosted MCP endpoint — nothing to install; the store header pins every session to this project and the host fixes the environment. The same MCP token card as on the SDK tab, with Download MCP package: a small folder holding only the connection — .env and .mcp.json with the project id and the token filled in, plus the agent rules AGENTS.md (and the CLAUDE.md pointer) covering the MCP tools and resources. Below it, step-by-step setup for Claude Code (.mcp.json, or claude mcp add once for every folder), Codex (a [mcp_servers] table in ~/.codex/config.toml) and Antigravity (mcp_config.json, Manage MCP Servers → View raw config), each config with a Copy button — for a manager the copy carries the real token, for everyone else a <token> placeholder to replace with a token from Access tokens. The package and the copied configs contain the token; keep them out of git.
Export#
Data Act export: every content collection per locale and the commercial data (orders, invoices, customers, products, form records) as downloadable files.