Tokens & access
Every knob is gated by a token. The scope of the token decides which loop you can turn.
Connect your agent
Point Claude Code (or any MCP client) at your site's endpoint:
claude mcp add loftur --transport http https://acme.loftur.app/mcp \
--header "Authorization: Bearer lft_your_owner_key"The endpoint also answers at loftur.app/mcp — it resolves your site from the key
alone, so you can connect before DNS propagates. The agent's first move should be
the site_help tool (the full authoring guide); the platform SKILL is at
loftur.app/skill.md.
Token types
| Token | Identity | Can do |
|---|---|---|
Owner key (lft_…) | one site | everything: code, schema (through the guard), content |
| Editor token | one site | content & assets only — no schema, no code |
Account PAT (lftr_pat_…) | an account (email) | claim/list/rotate sites + build any site you own |
- Owner key is minted when you claim a subdomain — a one-time secret. Lost it? Rotate from the dashboard.
- Editor token is minted from the dashboard (or
mint_editor_token). Give it to a content editor so they can maintain content over their own MCP client. It's an allowlist: content, publishing, and image tools only. Schema mutations and code tools aren't reachable — an editor can never even hit the migration guard. - Account PAT is one developer credential across all the sites you own. Every
build and content tool is re-exposed with a
siteselector, and each call is ownership-checked.
The editor scope is the enforcement seam that keeps the loops decoupled by person as well as by mechanism: content editors turn the content knob, never the schema or code knobs.
Which token for which loop
- Shipping content → editor token (or owner).
- Changing schema or code → owner key (or an account PAT for the site).
- Managing several sites as a developer → account PAT.