kdrive-mcp

MCP server for Infomaniak kDrive — search, browse, and read files directly from your drive inside any MCP-compatible AI client.

Features

• Search file content across the drive (full-text, indexed files)
• Find any file by name using parallel recursive directory traversal
• Browse the folder tree with list
• Read plain-text files (txt, csv, json, xml, md, …)
• Parse structured documents and return clean text:
  • Excel / Calc — .xlsx, .xls, .ods → CSV per sheet (SheetJS)
  • Word — .docx → plain text (mammoth)
  • PDF — .pdf → plain text (PDF.js via unpdf)
  • PowerPoint / Impress — .pptx, .ppt, .odp → plain text (officeparser)
  • Legacy / OpenDocument — .doc, .odt → plain text (officeparser)
• Smart router (kdrive_read) — auto-detects format, falls back to file extension when the MIME type is generic
• MCP Resource — expose any file as a base64 blob via kdrive://{id}

Tools

kdrive_search

Full-text search across indexed file content.

| Parameter | Type | Default | Description | |-----------|------|---------|-------------| | query | string | — | Search terms | | limit | integer | 15 | Max results (1–50) |

Note: Binary files (xlsx, pdf, …) are not indexed by kDrive's search engine. Use kdrive_find to locate them by filename.

kdrive_find

Recursively walks the directory tree (5 concurrent requests) to find files or folders by filename substring — reliable for all file types.

| Parameter | Type | Default | Description | |-----------|------|---------|-------------| | query | string | — | Filename substring (case-insensitive) | | folder_id | string | "1" | Folder to start from ("1" = drive root) | | max_results | integer | 10 | Max matches (1–50) |

kdrive_list

List the contents of a folder (files and subdirectories).

| Parameter | Type | Description | |-----------|------|-------------| | file_id | string | Folder ID to list ("1" = root) |

kdrive_read ⭐

Smart router — inspects the MIME type (with file-extension fallback) and delegates to the right parser automatically.

| Parameter | Type | Description | |-----------|------|-------------| | file_id | string | ID of the file to read |

Supported formats: xlsx, xls, ods, docx, doc, pdf, pptx, ppt, odp, odt, and any plain-text format.

kdrive_read_text_file

Read a plain-text file (txt, csv, json, xml, md, …) decoded as UTF-8. Output capped at 100,000 characters.

| Parameter | Type | Description | |-----------|------|-------------| | file_id | string | ID of the text file |

kdrive_read_excel

Parse a spreadsheet and return all sheets as CSV. Supports .xlsx, .xls, and .ods. Output capped at 80,000 characters.

| Parameter | Type | Description | |-----------|------|-------------| | file_id | string | ID of the spreadsheet file |

kdrive_read_word

Extract plain text from a .docx file using mammoth (highest quality). Output capped at 80,000 characters.

| Parameter | Type | Description | |-----------|------|-------------| | file_id | string | ID of the .docx file |

For legacy .doc files use kdrive_read.

kdrive_read_pdf

Extract plain text from a PDF using PDF.js (via unpdf). All pages merged. Output capped at 80,000 characters.

| Parameter | Type | Description | |-----------|------|-------------| | file_id | string | ID of the PDF file |

kdrive_read_pptx

Extract slide text from a presentation using officeparser. Supports .pptx, .ppt, and .odp. Output capped at 80,000 characters.

| Parameter | Type | Description | |-----------|------|-------------| | file_id | string | ID of the presentation file |

Resources

kdrive://{id}

Exposes any kDrive file as a base64-encoded blob resource with the correct MIME type — useful for GUI-driven file selection in MCP clients that support resources.

Setup

Prerequisites

You need two values from your Infomaniak account:

1. KDRIVE_TOKEN — an Infomaniak API token with kDrive access.
   Create one at app.infomaniak.com → API (scope: Drive).

2. KDRIVE_ID — the numeric ID of your kDrive.
   Find it in the kDrive web app URL: https://kdrive.infomaniak.com/app/drive/**123456**/…

Claude Desktop

Add this to your claude_desktop_config.json:

{
  "mcpServers": {
    "kdrive": {
      "command": "npx",
      "args": ["-y", "kdrive-mcp"],
      "env": {
        "KDRIVE_TOKEN": "your-api-token",
        "KDRIVE_ID": "your-drive-id"
      }
    }
  }
}

Config file locations:

• macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
• Windows: %APPDATA%\Claude\claude_desktop_config.json

Claude Code (CLI)

claude mcp add kdrive-mcp --scope project \
  -e KDRIVE_TOKEN=your-api-token \
  -e KDRIVE_ID=your-drive-id \
  -- npx -y kdrive-mcp

VS Code

Add to your workspace .vscode/mcp.json:

{
  "servers": {
    "kdrive": {
      "type": "stdio",
      "command": "npx",
      "args": ["-y", "kdrive-mcp"],
      "env": {
        "KDRIVE_TOKEN": "your-api-token",
        "KDRIVE_ID": "your-drive-id"
      }
    }
  }
}

Development

git clone https://github.com/ddanssaert/kdrive-mcp
cd kdrive-mcp
npm install

KDRIVE_TOKEN=your-token KDRIVE_ID=your-drive-id node index.js

Inspect tools interactively with the MCP Inspector:

KDRIVE_TOKEN=your-token KDRIVE_ID=your-drive-id \
  npx @modelcontextprotocol/inspector node index.js

Design Notes

• Context bloat prevention — search and list results are mapped to minimal objects (id, name, mime_type, size). Raw API responses with internal metadata are never forwarded to the model.
• Binary safety — all downloads use response.arrayBuffer(). Text is never decoded with response.text() + btoa(), which would corrupt non-UTF-8 bytes.
• Output caps — parsed document text is sliced to 80,000–100,000 characters with a truncation notice so the model always knows when content is incomplete.
• Timeout protection — API calls abort after 15 s; file downloads after 2 min.
• Rate limit awareness — kDrive enforces 60 requests/min. kdrive_find batches directory listings 5 at a time to stay efficient without bursting.

License

MIT