MCP tools for AI Data Engine governed AI access
A connected AI client can use a set of tools through the Model Context Protocol (MCP) server for AI Data Engine (AIDE) to search and tag files in a collection. Use this information to understand what each tool does and the access that the tools require.
Access requirements
AI clients use the tools in a power-user session. A user gets a power-user session only when both of these are true:
-
The user holds the
AIDEPowerUsersapp role in Microsoft Entra ID. -
The AI client requests the
power_usersscope.
A connection is limited to the collection in the MCP server URL. To work with a different collection, use the connection details for that collection.
Tools
| Tool | Description |
|---|---|
|
Returns the collection that your connection is limited to. |
|
Searches file metadata in a collection by name prefix, file type, or owner, with paging. In a power-user session, search isn't filtered by file-level permissions and returns every file in the collection that matches. Results include each file's name, path, type, size, last-modified time, and owner. |
|
Lists the tags that are defined for your tenant, and the number of files that have each tag. |
|
Creates a tag for your tenant and returns its ID and name. |
|
Adds a tag to a file or removes a tag from a file. If the tag doesn't exist, the tool creates it first. An AI client works on one file at a time. |
How tags work
-
Tags apply across AI Data Engine, not only to one collection. Tags that you create or apply through an AI client are also available in the Investigation view.
-
A misspelled tag name creates a new tag. Tag names are case-sensitive, so
Financeandfinanceare different tags. -
You can't delete or rename a tag after you create it. For limitations on tag names, refer to Known limitations.
-
AI clients can't list the files that have a tag.
What AI clients can't do
An AI client can't retrieve file contents, add files, delete files, modify files, change data sources, or create collections.