Skip to main content
Capell home
My account

We use cookies for essential features and optional analytics.

Manage cookie preferences

Choose which optional cookies Capell may use.

Essential
Always on

Required for Capell to work — login, security, page rendering.

Analytics

Anonymous usage telemetry to improve editor performance.

Marketing

Personalization for the Capell marketing site.

Cookie policy

Build a Capell website with an AI agent.

The read side needs no credentials: an agent can list capabilities, fetch the interview script and the site-spec JSON schema, and validate a draft spec against the public endpoint. Preview, export, and build require an Agent Bridge token. Capell performs no AI inference itself — your agent does the reasoning and the writing.

An AI agent reading Capell's structured site model — themes, page types, layouts — and assembling a validated site spec before build.

Capell exposes its site-design capabilities over MCP. An AI agent can discover your themes, page and section types, and layouts, fetch the interview script and the site-spec JSON schema, and validate a draft — all without credentials. Preview, export, and build authenticate against an Agent Bridge token and stay audited. Capell runs no inference of its own: your agent does the reasoning and the writing, and Capell builds the validated spec deterministically.

The flow

How an agent builds a Capell site

Discover

List installed themes, page types, section types, and layouts the spec may use.

Interview

Fetch the deterministic interview script and ask the user in the agent's own words.

Assemble

Build a site spec — pages, sections, theme, copy — against the published JSON schema.

Validate

Run validate_spec until the draft is valid. No data is written.

Preview

Authenticate, then build a non-destructive preview before any deploy.

Build

Confirm the preview to publish, or export a portable spec and install command.

Two MCP endpoints: a tokenless read surface for discovery and validation, and a token-gated surface for preview, export, and build.

Two endpoints: public read, authenticated build

  • Public read — ai/capell/read: list capabilities, get the interview, get the site-spec schema, and validate a spec. No authentication.
  • Authenticated — ai/capell: preview, export, build, and apply, behind an Agent Bridge token with per-token throttling.
  • The read endpoint runs a fixed allowlist of pure-read capabilities; the allowlist is the only trust boundary.
How AI assists editing too

The agent proposes. Capell stays deterministic.

The agent assembles a spec; Capell validates and builds it the same way every time. Mutations are token-gated and audited, and a preview always precedes an irreversible build.

No inference in Capell

Your agent reasons; Capell builds

Token-gated writes

preview / export / build / apply

Audited

every call recorded, even tokenless reads

Preview first

non-destructive before deploy

Questions

AI agent questions

Which agents can use this?
Any MCP-capable client. Capell speaks MCP over streamable HTTP.
Do I need an API key for read access?
No. The ai/capell/read endpoint is tokenless. Building requires an Agent Bridge token.
What can the public endpoint touch?
Only a fixed allowlist of read capabilities: catalogs, the interview, the spec schema, and spec validation. Nothing writes.
How does an agent build, then?
Authenticate against ai/capell, run a non-destructive preview, then publish it or export a portable spec.

Next step

Give an AI agent a CMS it can actually build on.

Capell's structured Laravel model (pages, URLs, types, layouts) is what makes agent-built sites predictable.

See the live demo, or read how Capell assists editing inside the admin once a site is built.

See the live demo AI-assisted editing
Loading footer