You can turn “attach the page that shows the problem” into one prompt. The GroupDocs.Viewer.Mcp server renders the pages you name from a PDF, Word, Excel or PowerPoint file and saves each as a PNG with a predictable file name, locally on your machine, so Claude, Cursor or GitHub Copilot can produce the images for a bug report, a review note or a slide:

Render pages 1, 4, and 9 of the specification and tell me where the files are.

The step-by-step version with config and troubleshooting is in the documentation: How to create page images of documents for reports and tickets.

Why automate page images at all?

A page image is the cheapest evidence there is. A support engineer attaches the invoice page that shows the wrong total; a reviewer drops a thumbnail of each changed page into the review notes; a presenter needs one diagram, unreformatted. Done by hand, each of those means opening the file, finding the page, taking a screenshot and naming the result. The page count differs per file, so the work never batches well.

An agent can do the repetitive part if the rendering is done by a real engine rather than by the agent imitating one. The engine paginates the document the way the document paginates and writes the same kind of PNG every time.

What is the workflow?

  1. Put the files where the server can see them. Mount the folder that holds the documents, whether it contains PDF files or Word files, so that rendering a DOCX page image works the same way. The server reads from the storage path and writes the images to the output path, which you can keep separate with GROUPDOCS_MCP_OUTPUT_PATH.
  2. Check the page count. Prompt: “How many pages does the specification have?” The agent calls get_view_info and reads pageCount. This matters because render_page does not reject a page number past the end of the document.
  3. Name the pages. Prompt: “Render pages 1, 4, and 9 of the specification and tell me where the files are.” The agent calls render_page three times, once per page.
  4. Collect the file names. The files are specification_page1.png, specification_page4.png and specification_page9.png, in the output folder. The agent reports the paths.
  5. Attach them. If your client also has an issue-tracker or chat MCP server connected, ask the agent to attach the files to the ticket or message. That step uses the other server’s tools, not this one.

If you only want the files, say so, so the agent does not describe every image:

Just save the images; you do not need to describe them.

What does the agent decide, and what does the engine do?

Decision or action Agent Engine
Which pages are relevant Maps “the pages with the pricing table” to page numbers, after rendering or inspecting Nothing
Whether the page exists Calls get_view_info and compares Returns pageCount and page sizes
Rendering Calls render_page once per page Produces the PNG
File naming Reports the saved path Saves as <source-stem>_page<N>.png
Attaching to a ticket Uses another server’s tool Nothing

Example session

An illustration based on the documented output names, not a captured screenshot.

You:   Render pages 1, 4, and 9 of the specification and tell me where
       the files are. Just save the images; no need to describe them.

Agent: [calls render_page for page 1, page 4 and page 9]
       Saved:
         <output folder>/specification_page1.png
         <output folder>/specification_page4.png
         <output folder>/specification_page9.png

How do I keep the set small?

Rendering is one page per call, and each image also travels back to the agent inline. Asking for ten pages means ten calls and ten images in the conversation. Ask for the pages you need. A sensible rule is to pick the pages before the render, using the page count and, for long documents, a contents page.

Typical uses

  • Bug reports. The page of the manual or the invoice that shows the problem.
  • Document reviews. One image per changed page in the review notes.
  • Slides. A picture of the one diagram you want, without reformatting it.

Honest limits

  • Evaluation mode. Rendered pages carry an evaluation watermark, and a server process opens at most 15 documents. If the images leave your team, ask “Is the viewer server licensed?” first, so the agent calls get_license_status.
  • One page per call. There is no multi-page tool. A batch is a series of calls.
  • PNG only. The format is fixed; resolution is not a parameter you can set through the tool.
  • A wrong page number still gives an image. A file called _page12.png is page 12 only if the document has 12 pages. Check the count first.

FAQ

How do I convert a PDF page to an image with an AI agent? Connect the GroupDocs.Viewer.Mcp server to your client and ask for the page by number. The agent calls render_page and the PNG is saved in the output folder as <name>_page<N>.png.

How do I attach a document preview to a ticket with an AI agent? The Viewer server produces the image file. Attaching it needs an issue-tracker MCP server connected to the same client, which is a separate server with its own tools.

Can I generate document thumbnails locally? Yes. The engine runs in the server process on your machine, and the images are written to the output folder you configure.

Go deeper