MCP server
AISiteEngine ships an MCP (Model Context Protocol) endpoint, so assistants and agents can read your content in a structured way instead of scraping the site.
What it exposes
Section titled “What it exposes”| Tool | Returns |
|---|---|
list_sites |
The sites the token can see, with record counts |
read_content |
Content records for a site, filtered by type/status |
read_design_tokens |
Design tokens extracted from the site |
Access is read-only and site-scoped: the token decides what the client can see.
Getting a token
Section titled “Getting a token”An administrator mints an MCP token from the dashboard’s API. It is a long-lived service token scoped to one site, intended for a specific integration — treat it like a password.
Connecting a client
Section titled “Connecting a client”The endpoint speaks MCP over HTTP at:
<mcp-base>/mcpwith the token in an Authorization: Bearer … header.
Example: a question
Section titled “Example: a question”Once connected, an assistant can answer questions like “what are the five newest published insights?” by calling
read_content with type=insights, status=publish, ordered by date — no crawling, no guessing at URLs.
Why not just the REST API?
Section titled “Why not just the REST API?”You can call the REST API directly. MCP exists so that AI clients get a typed, discoverable surface with descriptions attached, instead of a generic set of endpoints they have to be told about in a prompt.