How to audit document metadata with an AI agent
Leave feedback
On this page
Files carry more than their content: author names, machine names, company fields, editing history, camera data, GPS coordinates. Reading it is one tool call — and knowing what to ask is the whole skill.
Note
The commands and config snippets on this page are for the .NET build of the server — the only platform available today. Installation and client setup: MCP server for .NET. Other platforms will expose the same tools with their own launch command; everything else on this page applies unchanged.
The pattern
Put the files in the storage folder the server can see.
Ask: “What metadata does report.pdf carry?”
The agent calls read_metadata and reports the properties grouped by package.
Follow up in plain language; the agent already holds the data.
Ask precisely, get more
Broad reads produce long lists. Targeted questions produce answers:
Does any property in these files still contain the old company name?
Which of these photos have GPS coordinates?
Who is listed as last-modified-by across the folder?
Show me only the custom properties.
The second and third are search_metadata with a valueContains or nameContains filter — one call per file, and only the matches come back.
The finding people are usually looking for
Three properties account for most privacy surprises:
Author / last-modified-by — often a real person’s name, sometimes a former employee’s.
Company — frequently a previous employer, inherited through a template.
GPS in EXIF — a photo taken at home, shared publicly.
Ask for those three by name and you have covered the common cases in one pass.
The licence caveat that matters here more than anywhere
Unlicensed, the engine returns only the first five document properties, and neither the response nor the agent mentions it. An audit that reports “nothing sensitive” may simply have stopped reading. Before you trust a result:
What is the license status of the metadata server?