# convertisseur-ia [Health: Active]

**Category:** 🏢 Workplace & Productivity  
**Repository:** https://github.com/Walmes1/connecteurs-ia  
**GitHub Stars:** 0  
**npm Downloads (last month):** 110  
**Views:** 2  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/convertisseur-ia

## Description
Documents (PDF, Word, Excel, PowerPoint, CSV) vers Markdown leger pour l'IA, 100 % local

## Claude Desktop Quick Installation
Install path detected from listing signals. Uses `npx` (confidence: high):

```json
"mcpServers": {
  "convertisseur-ia": {
    "command": "npx",
    "args": ["-y","convertisseur-ia-mcp"]
  }
}
```

## Documentation

## What convertisseur-ia MCP server does

The convertisseur-ia MCP server reads files from the local filesystem and turns supported documents into compact Markdown for use in an MCP conversation. It is intended for documents that already contain a text layer or structured text, rather than scanned pages that require optical character recognition. Supported inputs include text-based PDF files, DOCX, XLSX, PPTX, CSV, HTML, JSON, XML, RTF, TXT, and source code.

The conversion process also estimates token usage. This lets an agent compare the approximate size of the original content with the lighter Markdown result before sending more material into a conversation. The server operates locally: files are read from disk, converted in memory, and no network request is made.

## How it works

The server runs as a Node.js MCP process using standard input and output. For Office Open XML files such as DOCX, XLSX, and PPTX, it reads the ZIP package and converts the contained XML into Markdown without requiring a separate Office installation. PDF text is extracted with pdf.js, with an integrated fallback when that library is unavailable.

PDF files are checked for a usable text layer. If a PDF is identified as scanned, the conversion tool reports that it should be handled by the separate `ocr-images` server. That companion server uses Windows OCR and is not part of the document conversion tools described here.

Token figures are estimates rather than billing measurements. Text is approximated at about four characters per token, while image estimates use width multiplied by height divided by 750.

## Setup and configuration

Node.js 18 or newer is required, and the converter is described as cross-platform. The published package can be started with `npx -y convertisseur-ia-mcp`. A source checkout can instead be installed with `npm install` from the `convertisseur-ia` directory.

Add the process to an MCP client's `mcpServers` configuration with `npx` as the command and `-y convertisseur-ia-mcp` as its arguments. The client must be restarted after changing its configuration. The project names Claude Desktop, Claude Code, Cursor, VS Code, and LM Studio as clients that can be configured this way.

## Tools and capabilities

The convertisseur-ia MCP server exposes five tools:

- `convertir_document` converts one file to Markdown and returns token statistics. It accepts character limits and optional starting and ending pages.
- `apercu_document` returns a low-cost preview containing the detected type, file size, estimated tokens, and initial characters.
- `convertir_dossier` combines a directory's files into one Markdown result. Filtering, recursion, and a maximum file count can be specified.
- `enregistrer_markdown` converts a file and writes a `.md` file beside the source by default.
- `formats_supportes` lists recognized formats and relevant exceptions.

## Limitations and notes

The convertisseur-ia MCP server reads any local file path that the trusted MCP client asks it to read; it does not impose a directory restriction. It should therefore be exposed only to a trusted client and not placed behind a public service.

Only `enregistrer_markdown` writes output files among these tools. Older binary Office formats such as DOC, XLS, and PPT are rejected and need to be saved as DOCX, XLSX, or PPTX first. Scanned PDFs need the separate OCR workflow. The MIT-licensed project also notes that conversion output and token counts should be treated as estimates for practical use.

_Full upstream README: https://allmcps.com/mcp/convertisseur-ia/readme_