How to convert documents with AI agents using MCP Leave feedback

Use any MCP-compatible AI agent — Claude, Cursor, GitHub Copilot, Windsurf, Codex — to convert documents between PDF, Word, Excel, PowerPoint, Markdown, and 100+ other formats, locally. Install the GroupDocs.Conversion MCP server with one command, then ask in plain language:

dnx GroupDocs.Conversion.Mcp --yes

Convert contract.docx to PDF

This page shows the pattern once; every specific scenario links from here.

AI agents are excellent at deciding what to do with documents — and unreliable at doing it. Rendering fidelity, fonts, tables, page geometry: that is deterministic engine work, not generation work. The MCP pattern puts each side where it belongs:

• The agent interprets your intent, picks files, chooses tools, sequences steps, handles results.
• The engine (GroupDocs.Conversion — the same one behind the .NET/Java libraries) performs the conversion with production fidelity.

The server exposes three tools — convert, get_supported_formats, get_document_info — and your prompts drive them.

Register the server in your client — full per-client reference; Claude Desktop example:

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

Restart the client. Documents go in the storage folder; conversions come back there (or in your configured output folder).

PDF → DOCX (make a PDF editable):

Convert scan-agreement.pdf to DOCX so I can edit it

DOCX → PDF (finalize for distribution):

Convert the final-report.docx to PDF

Both run through the same convert tool; the agent fills in the target format. Open-ended questions work too — “What formats can I turn this Visio file into?” calls get_supported_formats, and “Check the page count first, then convert” chains get_document_info before converting.

• PDF → Markdown for RAG / knowledge bases
• Markdown → PDF / Word (export what the agent wrote)
• Batch-convert an entire folder
• On-premise / regulated deployment
• All supported formats

Is this conversion or extraction? Conversion — full-fidelity format transformation. For field-level data extraction (values, structured tables, text spans), use the companion GroupDocs.Parser MCP server; many pipelines use both.

Do my files get uploaded anywhere? No. The server runs locally over stdio; the data path is agent → local server → local filesystem. See on-premise architecture.

Which AI clients work? Claude Desktop, Claude Code, VS Code / GitHub Copilot, Visual Studio 2022, Cursor, Windsurf, Cline, Codex CLI, JetBrains Rider — setup for each.