MarkPrep
All posts
Published · Updated · 6 min

Let Your AI Assistant Convert Documents Itself (MCP)

An MCP server lets an AI assistant such as Claude convert a document to Markdown itself, from a file path you give it. Without one, you have a PDF and want to ask Claude about it. You leave the conversation, find a converter, turn the PDF into Markdown, come back, paste it in, and then ask your question. The assistant can read the pasted text, but it can't open the file on your disk, so every conversion is a job for you.

What MCP is

MCP, the Model Context Protocol, is a standard way to give an AI assistant new tools. An MCP server runs on your machine and offers actions the assistant can call directly. Clients such as Claude Code and Claude Desktop speak the protocol.

A document-conversion MCP server offers one action: "convert this file to Markdown." Once it is connected, your assistant can pass a file path and get Markdown back without you doing anything. If you have looked for a "convert documents to Markdown skill" for Claude, an MCP server covers that need. You set it up once in the client, and the assistant calls it as a tool whenever a task needs a file converted.

The workflow, before and after

Before: open a converter, convert the file, copy the Markdown, return to the chat, paste, ask.

After: "Convert the report on my desktop and summarise the risks section." The assistant calls the tool, gets the Markdown and answers, all from one message.

It converts on your machine

Letting an assistant convert your files is only comfortable if the files aren't sent anywhere. MarkPrep's MCP server runs the conversion on your machine, with a local Node converter built on the same core libraries as the web app. The file the assistant converts never leaves your device, so a contract or an internal report stays with you. The server does need an internet connection, because it checks your licence key online on each run. That check is its only network call.

The MCP server and the command-line converter handle .pdf, .docx, .xlsx, .xls, .html, .htm, .csv, .tsv, .json, .md, .markdown and .txt files. They have no OCR and don't read PowerPoint or EPUB files; use the web app for those. PDFs come out as plain lines of text, one page after another, without the headings and tables the web app rebuilds.

Setting it up

The MCP server comes with MarkPrep Pro, a one-time £19 purchase that gives you a licence key. In Claude Code, add the server with one command:

claude mcp add --transport stdio --env MARKPREP_LICENSE=your-licence-key markprep -- npx -y markprep mcp

For Claude Desktop, or a project's .mcp.json file in Claude Code, add the same server to the config file:

{
  "mcpServers": {
    "markprep": {
      "command": "npx",
      "args": ["-y", "markprep", "mcp"],
      "env": { "MARKPREP_LICENSE": "your-licence-key" }
    }
  }
}

Restart the client and your assistant has a convert_document tool, which takes the file's absolute path. The same package gives you the command-line converter for scripts and terminals, such as npx markprep convert report.pdf -o report.md. It also has a watch mode, npx markprep watch ./inbox -o ./markdown, which converts any file dropped into a folder. Full setup is on the MCP page.

Who this is for

If you paste documents into an assistant a few times a month, the manual way is fine. If you feed it contracts, reports or research all the time, the server saves you converting, copying and pasting each file.

Follow the setup on the MCP page to connect the server, then ask your assistant to convert a file by its path. For batch jobs and watch folders without an assistant, see the command-line converter.

Try it on your own file

Convert a document and watch the token counter. It's free, needs no account and uploads nothing.