How to batch-convert a folder of documents with an AI agent Leave feedback

Convert an entire folder of documents with a single prompt — the agent orchestrates, the GroupDocs engine batch-converts locally; nothing is uploaded and your originals are kept. Point the server at the folder and ask:

Convert every DOCX in my documents folder to PDF, keep the originals

The agent enumerates the files, calls convert per document, and reports what it produced. No uploads, no per-file clicking, no script to maintain.

Batch scenarios live or die on the folder mapping. Set it once and the whole folder is in scope for every prompt:

{
  "mcpServers": {
    "groupdocs-conversion": {
      "type": "stdio",
      "command": "dnx",
      "args": ["GroupDocs.Conversion.Mcp", "--yes"],
      "env": {
        "GROUPDOCS_MCP_STORAGE_PATH": "/path/to/invoices",
        "GROUPDOCS_MCP_OUTPUT_PATH": "/path/to/invoices/converted"
      }
    }
  }
}

With a separate GROUPDOCS_MCP_OUTPUT_PATH, originals stay untouched in storage and every converted file lands in the output folder — the cleanest batch layout. (Installer users: this is storagePath / outputPath in one config.)

• “Convert everything in this folder from DOCX to PDF.”
• “Turn all the PDFs into Markdown; skip files that are already converted.”
• “Convert the invoices to PDF/A for archiving, and list any file that failed.”
• “Make PNG previews of every presentation.”

The prompt is the orchestration layer: filtering (“only the 2025 ones”), sequencing (“check page counts first”), and error policy (“list failures at the end”) are all plain language — the deterministic conversion stays in the engine.

• Existing files are never overwritten — re-running a batch produces distinctly-named files rather than clobbering earlier output.
• Large batches: agents typically process files sequentially; a few hundred documents work fine, and the report at the end tells you exactly what happened. For scheduled/unattended pipelines beyond agent sessions, the same engine is scriptable via the .NET library.
• Evaluation mode caps a single server process at 15 opened documents — a license lifts this for real batches.

Can Claude convert an entire folder at once? Yes — one prompt; the agent iterates the folder through the local MCP server. You never upload files one by one.

How do I keep originals? They are kept by default — conversion writes new files. A separate output folder (above) keeps the source folder pristine.

Can I filter which files get converted? Say it in the prompt — by extension, name pattern, or even content questions the agent checks via get_document_info.