The scenario
You are working in a client outside this website — an assistant, an editor, a script — and you want this site's content available to it as data.
The MCP server exposes Northmoor's content as tools. The same information the search demo assembles from prose becomes something an external agent can query directly and reason over.
Connect
Point your MCP client at this site's server endpoint. The connection details are on the site's integration settings; a client that supports MCP will discover the available tools automatically on connect.
What is exposed
Content discovery. List and filter content by type. Programs by degree level, courses by code prefix, people by role, events by date range.
Entity retrieval. Fetch a specific entity and its fields, including the relationships stored on it.
Relationship traversal. Follow references in the direction they are stored. Department to faculty, program to courses, news to the people and programs it tags.
Search. The same retrieval the on-site search uses, available as a tool.
One action. Event registration, as the example of a write operation.
Try this
Using the Northmoor tools, list every graduate program and tell me which department offers each one.
This is more interesting than it looks. Programs hold no reference to departments — the relationship is stored on the department, pointing outward. An agent that only reads the program entities will not find the answer. It has to enumerate departments and look at what each one lists.
What you should see. Five graduate programs, correctly attributed, with Coastal Resilience appearing under two departments because it is genuinely delivered by both.
Then try
Which courses does Dr. Miriam Frost teach?
Person entities store nothing. The teaching relationship lives on the course, in course_instructors. A tool call that fetches the person and reports her fields will return no courses at all. Getting this right means querying courses filtered by instructor.
This is the same modelling decision the site's own pages rely on, and it is the most common thing an integration gets wrong.
Why this matters
An MCP server turns a website into an interface for agents. The content model that makes the site coherent for readers is the same one that makes it queryable for machines — which is an argument for modelling relationships properly rather than duplicating them.