Command Palette

Search for a command to run...

UnylyUnyly
Весь каталог

Doc Sanitizer Server

БесплатноНе проверен

A local, containerized MCP server that uses a local LLM to sanitize documents by removing or transforming PII before content is sent to public LLM services.

GitHubEmbed

Описание

A local, containerized MCP server that uses a local LLM to sanitize documents by removing or transforming PII before content is sent to public LLM services.

README

A local, containerized MCP (Model Context Protocol) server that uses a local LLM (via Ollama) to sanitize documents by removing or transforming PII before content is sent to public LLM services.

Features

  • Document Support: Word (.docx), PDF, Excel (.xlsx/.xls), CSV, Plain text, Email (.eml)
  • PII Detection: Names, emails, phone numbers, companies, addresses, financial data, IDs
  • Flexible Actions: Delete, invent synthetic data, or keep partial information
  • Profile Management: Create and manage multiple sanitization profiles via CLI
  • MCP Integration: Works with Claude Desktop, Claude.ai, and Claude Code
  • Privacy First: All processing happens locally - nothing sent to external services

Quick Start

Prerequisites

  • Docker and Docker Compose
  • Ollama (installed automatically via Docker)

Installation

git clone https://github.com/yourusername/doc-sanitizer-mcp.git
cd doc-sanitizer-mcp
./setup.sh

Configuration

Edit .env to customize settings:

# Change LLM model (no rebuild required)
OLLAMA_MODEL=phi4:14b

# Alternative models:
# OLLAMA_MODEL=qwen2.5:7b    # Faster, less RAM
# OLLAMA_MODEL=llama3.2:3b   # Fastest, minimal RAM

To switch models:

# Pull new model
docker exec doc-sanitizer-ollama ollama pull qwen2.5:7b

# Update .env
sed -i '' 's/OLLAMA_MODEL=.*/OLLAMA_MODEL=qwen2.5:7b/' .env

# Restart
docker compose restart doc-sanitizer

Claude Desktop Integration

Add to ~/Library/Application Support/Claude/claude_desktop_config.json:

{
  "mcpServers": {
    "doc-sanitizer": {
      "url": "http://localhost:8000/sse"
    }
  }
}

Usage

Via Claude (MCP Tools)

Once configured, you can ask Claude:

  • "List available sanitization profiles"
  • "Show me the default profile settings"
  • "Sanitize this document using the high_privacy profile"

Via CLI

# List profiles
doc-sanitizer profiles list

# Create a new profile
doc-sanitizer profiles create high_privacy

# Edit profile settings
doc-sanitizer profiles edit high_privacy --set person_name=delete --set company=delete

# Sanitize a document
doc-sanitizer sanitize document.docx --profile high_privacy

Profile Management

Profiles define how each PII type is handled:

PII Type Available Actions
person_name delete, invent, keep_part
email delete, keep_part
phone delete, invent, keep_part
company keep_part, invent
address delete, invent
financial delete, invent
id_numbers delete, invent
date_of_birth delete, invent

Profiles persist across container restarts (stored in ~/.doc-sanitizer/).

Documentation

License

MIT License - see LICENSE for details.

from github.com/yannickgiguere-dfs/doc-sanitizer-mcp

Установка Doc Sanitizer Server

У этого сервера нет опубликованного пакета — он собирается из исходников. Открой репозиторий и следуй инструкции в README.

▸ github.com/yannickgiguere-dfs/doc-sanitizer-mcp

FAQ

Doc Sanitizer Server MCP бесплатный?

Да, Doc Sanitizer Server MCP бесплатный — установка в пару кликов через Unyly без оплаты.

Нужен ли API-ключ для Doc Sanitizer Server?

Нет, Doc Sanitizer Server работает без API-ключей и переменных окружения.

Doc Sanitizer Server — hosted или self-hosted?

Self-hosted: сервер запускается локально на твоей машине командой из раздела установки.

Как установить Doc Sanitizer Server в Claude Desktop, Claude Code или Cursor?

Открой Doc Sanitizer Server на unyly.org, выбери вкладку своего клиента (Claude Desktop, Claude Code, Cursor) и нажми Install — конфиг сгенерируется автоматически, без правки JSON.

Похожие MCP

Compare Doc Sanitizer Server with

Не уверен что выбрать?

Найди свой стек за 60 секунд

Автор?

Embed-бейдж для README

Похожее

Все в категории ai