Show HN: MCP Server for Document Processing via Natural Language

4 days ago 1

A Model Context Protocol (MCP) server implementation that integrates with the Nutrient Document Web Service (DWS) Processor API, providing powerful PDF processing capabilities for AI assistants.

This server allows AI assistants to access the tools provided by Nutrient DWS Processor API, enabling operations such as digital signing, document generation, document editing, OCR, watermarking, redaction, and more.

Features Overview
Usage
Contributions

Feature Description

Document Creation	Merge PDFs, Office docs, and images
Editing	Watermark, rotate, flatten, redact, and more
Format Conversion	PDF ⇄ DOCX, images, PDF/A support
Digital Signing	Add PAdES standards-compliant digital signatures using trusted certificates
Data Extraction	Extract text, tables, or structured content
Security	Redaction presets, password protection, permission control
Advanced OCR	Multi-language, image and scan recognition
Optimization	Compress files without quality loss

Getting Started with Claude Desktop + Nutrient DWS MCP Server

Get a Nutrient DWS API key: Sign up at nutrient.io/api.
Download Claude Desktop: If you haven’t already, download Claude Desktop and sign in.
Install Node.js: Install Node.js with a package manager like brew on the command line. (brew install node)
Configure Claude: Add your API key and set the directory you’d like to use as a sandbox in claude_desktop_config.json (~/Library/Application\ Support/Claude/claude_desktop_config.json on macOS).

{ "mcpServers": { "nutrient-dws": { "command": "npx", "args": ["-y", "@nutrient-sdk/dws-mcp-server", "--sandbox", "/your/sandbox/directory"], "env": { "NUTRIENT_DWS_API_KEY": "YOUR_API_KEY_HERE" } } } }

Restart Claude Desktop.
Add documents for processing: Use any file manager to copy the documents into the sandbox directory set in the claude_desktop_config.json file above.
Process documents: Instruct Claude Sonnet 3.7 (e.g. “redact all PII from secret.pdf”, “sign the document contract.pdf”, “merge secret.pdf and contract.pdf together”, etc.).

Note: All operations involve reading from and writing to files on disk. We strongly recommend using the sandboxed directory feature to enhance security and prevent data loss.

Nutrient DWS MCP Server has been tested with Claude Desktop (Claude 3.7 Sonnet). Other MCP clients may work, but results may vary.

Nutrient DWS MCP Server only supports macOS for now. Feel free to open an issue if you’re interested in Windows/Linux support.

Sandbox mode (Recommended)

The server supports an optional sandbox mode that restricts file operations to a specific directory. This is useful for security purposes, ensuring that the server can only read from and write to files within the specified directory. You should drop any documents you'd like to work on in this directory.

To enable sandbox mode, use the --sandbox (or -s) command line argument followed by the path to the directory:

npx @nutrient-sdk/dws-mcp-server --sandbox /path/to/sandbox/directory

When sandbox mode is enabled:

For relative paths, they are resolved relative to the sandbox directory.
All input file paths are validated to ensure they exist and reside in the sandbox before performing any file operations

If no sandbox directory is specified, the server will operate without file path restrictions, allowing access to any file on the system that the server process has permission to access. (Not Recommended)

Processed files will be saved to a location determined by the LLM. If sandbox mode is enabled, it will reside inside this directory.

To further guide the LLM on where to place the output file, use natural language such as "please output the result to output/my_result.pdf". Or you may also add an output directory in your sandbox to hint to the LLM to use this directory for all resulting files.

Please see the contribution guidelines in CONTRIBUTING.md

Read Entire Article