How to convert documents to Markdown for RAG with an AI agent Leave feedback

Retrieval pipelines start with a conversion step, and that step decides how good the retrieval will be. Markdown is the format most chunkers and embedders handle best: headings survive, tables stay tables, lists stay lists.

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.

1. Put the corpus in the storage folder the server can see.
2. Ask: “Convert every PDF in this folder to Markdown, images as files, with front matter.”
3. The agent calls convert_to_markdown per document.
4. .md files (and an image folder) land in your output folder, ready to chunk.

The default embeds images as base64 data URIs, which makes a self-contained file — and makes a terrible RAG corpus: a single picture becomes a multi-megabyte blob sitting in the middle of a chunk, and your embedder gets a page of base64 instead of a paragraph of text.

For an index, use images: "file" (pictures written alongside, referenced by path) or skip images entirely when only the prose matters.

frontMatter: true writes YAML with title, author, format, and page count at the top of each file. That is chunk metadata you would otherwise have to reconstruct from file names — and it lets a retrieval answer cite “the Employee Handbook, page 42” instead of “handbook-final-v3.md”.

Evaluation mode converts only the first three pages. Nothing errors, nothing warns; you get a valid .md that contains the first three pages of a 200-page manual. Build an index from it and every retrieval quietly fails to find what it should.

Make the check the first step of any run:

Before converting anything: what is the license status of the markdown server?

get_license_status; see Licensing.

For each converted file, tell me the source page count and the resulting Markdown length.

A 300-page manual that became 4 KB of Markdown is either a scan with no text layer or a truncated run. Both are invisible unless you look.

dnx GroupDocs.Markdown.Mcp --yes

with GROUPDOCS_MCP_STORAGE_PATH pointing at the corpus — per-client config or the installer.

• Publish documents to a static site — front matter, images, and flavour.
• Convert a section, not a book — page ranges and worksheets.
• Migrate legacy documentation — CHM, EPUB, and old Word archives.
• On-premise architecture — the corpus never leaves.