document-loader-mcp

Document Loader MCP

Read PDF, Word, Excel, PowerPoint, and image files directly in Claude Code.

Source: AWS Labs - Document Loader MCP

What It Does

Tool	File Types	Description
read_document	PDF, Word, Excel, PowerPoint	Extract text content from documents
read_image	PNG, JPG, GIF, BMP, TIFF, WEBP	Load images for visual analysis

Installation

Prerequisites

• Node.js 18+ or Bun
• Claude Code CLI

Step 1: Install the MCP Package

# Using npm
npm install -g @anthropic/mcp-document-loader

# Or using bun
bun add -g @anthropic/mcp-document-loader

Step 2: Add to Claude Code Settings

Edit your Claude Code settings file:

# macOS/Linux
~/.claude/settings.json

# Or open via Claude Code
claude config

Add this to the mcpServers section:

{
  "mcpServers": {
    "document-loader-mcp": {
      "command": "npx",
      "args": ["-y", "@anthropic/mcp-document-loader"],
      "env": {}
    }
  }
}

Alternative using Bun:

{
  "mcpServers": {
    "document-loader-mcp": {
      "command": "bunx",
      "args": ["@anthropic/mcp-document-loader"],
      "env": {}
    }
  }
}

Step 3: Verify Installation

# List available MCPs
mcp-cli servers

# Check the tools
mcp-cli tools document-loader-mcp

# Should show:
# document-loader-mcp/read_document
# document-loader-mcp/read_image

Usage

Reading Documents

# Check the schema first (REQUIRED)
mcp-cli info document-loader-mcp/read_document

# Read a PDF
mcp-cli call document-loader-mcp/read_document '{"file_path": "/path/to/file.pdf", "file_type": "pdf"}'

# Read a Word document
mcp-cli call document-loader-mcp/read_document '{"file_path": "/path/to/file.docx", "file_type": "docx"}'

# Read an Excel file
mcp-cli call document-loader-mcp/read_document '{"file_path": "/path/to/file.xlsx", "file_type": "xlsx"}'

# Read a PowerPoint
mcp-cli call document-loader-mcp/read_document '{"file_path": "/path/to/file.pptx", "file_type": "pptx"}'

Reading Images

# Check the schema first
mcp-cli info document-loader-mcp/read_image

# Read an image
mcp-cli call document-loader-mcp/read_image '{"file_path": "/path/to/image.png"}'

Parameters

read_document

Parameter	Type	Required	Description
file_path	string	Yes	Absolute path to the document
file_type	string	Yes	One of: pdf, docx, doc, xlsx, xls, pptx, ppt
timeout_seconds	integer	No	Timeout (default: 30, min: 5, max: 300)

read_image

Parameter	Type	Required	Description
file_path	string	Yes	Absolute path to the image
timeout_seconds	integer	No	Timeout (default: 30, min: 5, max: 300)

Auto-Detection in CLAUDE.md

Add this to your ~/.claude/CLAUDE.md to make Claude automatically use this MCP:

### Document Reading - Use `document-loader-mcp` (NOT built-in Read tool)

| File Extension | MCP Command |
|----------------|-------------|
| `.pdf` | `mcp-cli call document-loader-mcp/read_document '{"file_path": "/path/to/file.pdf", "file_type": "pdf"}'` |
| `.doc`, `.docx` | `mcp-cli call document-loader-mcp/read_document '{"file_path": "/path/to/file.docx", "file_type": "docx"}'` |
| `.ppt`, `.pptx` | `mcp-cli call document-loader-mcp/read_document '{"file_path": "/path/to/file.pptx", "file_type": "pptx"}'` |
| `.xls`, `.xlsx` | `mcp-cli call document-loader-mcp/read_document '{"file_path": "/path/to/file.xlsx", "file_type": "xlsx"}'` |

**Auto-detection rules:**
1. File path ends in .pdf, .doc, .docx, .ppt, .pptx, .xls, .xlsx → USE document-loader-mcp IMMEDIATELY
2. File path ends in .png, .jpg, .jpeg, .gif, .bmp, .tiff, .webp → USE document-loader-mcp/read_image
3. **NEVER ask "should I read this?"** - just read it automatically

Troubleshooting

"MCP not found"

# Reinstall
npm install -g @anthropic/mcp-document-loader

# Or try with npx directly
npx -y @anthropic/mcp-document-loader

"Timeout" errors

Increase the timeout for large files:

mcp-cli call document-loader-mcp/read_document '{"file_path": "/path/to/large.pdf", "file_type": "pdf", "timeout_seconds": 120}'

"File not found"

Always use absolute paths, not relative:

# Wrong
mcp-cli call document-loader-mcp/read_document '{"file_path": "file.pdf", "file_type": "pdf"}'

# Correct
mcp-cli call document-loader-mcp/read_document '{"file_path": "/Users/you/Documents/file.pdf", "file_type": "pdf"}'

Why Use This Instead of Built-in Read?

Feature	Built-in Read	document-loader-mcp
PDF extraction	Basic	Full text + structure
Word documents	No	Yes
Excel files	No	Yes
PowerPoint	No	Yes
Image analysis	Yes	Yes
Large files	Limited	Better handling

License

Apache 2.0 (AWS Labs)