How to let an AI agent see document pages
Leave feedback
On this page
An agent that can read text still cannot see a chart, a stamp, a layout, or a scanned page. render_page turns any page into a PNG image and hands it to the agent, so a vision-capable model looks at the page the way you would.
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 document in the storage folder the server can see.
Ask: “Show me page 3 of quarterly.pptx and tell me what the chart says.”
The agent calls get_view_info to learn the page count, then render_page for the page it needs.
The image comes back inline; a copy is saved as quarterly_page3.png.
Always start with the page count
render_page does not check the page number. Ask for page 40 of a 12-page file and you still get an image — of a page that is not page 40. A good prompt makes the check explicit:
How many pages does the report have? Then show me the last one.
Pick the pages, not the document
Each page travels back as an inline PNG, typically a few hundred kilobytes. Rendering a 60-page document fills the agent’s context with images it does not need. Ask for the pages that answer the question, or have the agent render a few and decide from there.
Make sure the model can see
The image is only useful to a model that accepts image input. With a text-only model the agent still gets the saved file path, so it can pass the PNG on, but it cannot tell you what is on the page.
The licence check
In evaluation mode every rendered page carries a watermark, and one server process opens at most 15 documents. Before rendering pages you will share: