Share over MCP

Control what the ontology MCP can see

Once an ontology is exposed over MCP, the rest of the MCP Server page controls what a connected assistant can learn and return. Everything starts locked down; you open it up deliberately.

In the app: Data Context › Ontologies › your ontology › More › MCP Server

Before you start

  • The MCP server must be switched on for the ontology, and you need the Read/Write role. See Expose an ontology over MCP.
  • Changes save automatically.

Discovery

The four checkboxes under Discovery each widen what an assistant can learn before it asks a question. All start off: a new endpoint answers questions but reveals nothing else about the ontology.

Option What it allows
Concept search Search the ontology’s concepts by name and description before asking.
Glossary lookups Let the server consult your business glossary when it interprets a question.
Schema explorer Show physical table and column names, and turn on the binding-map tools. Leave it off and an assistant learns your business vocabulary but never your physical schema.
Named instances Let the server list the ontology’s named individuals.

Data MCP tools lists which tools each option turns on.

Never return raw values

Open Never return raw values to protect sensitive classes and attributes, such as a customer’s email address or a national ID.

  1. Choose Classes or Attributes.
  2. Search by name or description if the list is long.
  3. Tick each one to protect.
  4. Set its level in the Access column.
Level What it means
Never accessible The value can’t be selected, filtered, or aggregated at all. Use this for personal data.
Totals only The value can be used in totals, averages, and counts over the whole result, and in IS NULL / IS NOT NULL checks. It can’t be grouped by other columns or filtered in any other way, and individual values are never returned.

Redaction applies in every mode. Even with Include data, a protected value never comes back to the assistant.

Warning: A total over a single row equals that row’s value. Use Never accessible, not Totals only, for anything personal.

[screenshot]
The Never return raw values list with the Access column set to Never accessible and Totals only.

Per-user access

Open Per-user access to tighten the endpoint for individual people in your organization. Each row shows the person, their Role, a Data access choice, a Row limit, and an Effective badge with the mode and row limit they actually get.

  • Data access is either Endpoint default or Force SQL-only.
  • Row limit can only be lower than the endpoint’s Maximum rows per answer.

A per-person setting can only make the endpoint stricter for that person. There’s no way to give anyone data access, or a higher limit, than the endpoint itself allows.

Recent activity

Open Recent activity to see every call the server has handled, including the ones it refused. The columns are When, Tool, User, Question, Mode, Outcome, Rows, and Duration. Use the refresh button to load the latest calls.

[screenshot]
The Recent activity grid showing answered and refused calls.

Good practice

  • Start with SQL and explanation only and check the queries look right before switching to Include data.
  • Protect sensitive classes and attributes before you switch on Include data, not after.
  • Turn on Schema explorer only for assistants that need to see physical table names.
  • Look at Recent activity after the first week to see what people are actually asking.