Mcp Proxy For Pii Pseudo Anonymization

Name: Mcp Proxy For Pii Pseudo Anonymization
Author: gbrigandi

@gbrigandi

代码仓库

关于 Mcp Proxy For Pii Pseudo Anonymization

暂无概览

#mcp-proxy #pii-pseudo-anonymization #data-security

基本信息

分类

其他

传输方式

stdio

发布者

gbrigandi

提交者

gbrigandi

配置

使用下面的配置,将此服务器添加到你的 MCP 客户端。

{
  "mcpServers": {
    "conceal-database": {
      "command": "mcp-server-conceal",
      "args": [
        "--target-command",
        "python3",
        "--target-args",
        "database-server.py --host localhost",
        "--config",
        "/path/to/mcp-server-conceal.toml"
      ],
      "env": {
        "DATABASE_URL": "postgresql://localhost/mydb"
      }
    }
  }
}

工具

未检测到工具

工具是从 README 中自动提取的。维护者可以在 ## Tools 标题下列出工具,即可填充这部分内容。

概览

What is Mcp Proxy For Pii Pseudo Anonymization?

An MCP proxy that pseudo-anonymizes personally identifiable information (PII) before data reaches external AI providers like Claude, ChatGPT, or Gemini. It detects PII in responses from your MCP server and replaces it with realistic, structurally preserved fake data, maintaining semantic meaning and data relationships for AI analysis.

How to use Mcp Proxy For Pii Pseudo Anonymization?

Download or build the binary (mcp-server-conceal), install Ollama if using LLM-based detection modes, create a TOML configuration file, then run it as a proxy with --target-command and --target-args pointing to your existing MCP server. For example: mcp-server-conceal --target-command python3 --target-args "my-mcp-server.py" --config mcp-server-conceal.toml.

Key features of Mcp Proxy For Pii Pseudo Anonymization

Three detection modes: regex only, LLM only, or combined regex+LLM
Configurable regex patterns for emails, phones, SSNs, credit cards, IPs, URLs
Faker-based generation with locale and seed for consistent pseudo-anonymization
SQLite mapping database with configurable retention days
LLM detection result caching for performance
Supports custom prompt templates for domain-specific PII

Use cases of Mcp Proxy For Pii Pseudo Anonymization

Proxying MCP servers used with Claude Desktop to protect user PII
Sanitizing database or API responses before sending to external AI assistants
Adding a privacy layer to existing MCP tools without modifying their code
Production environments requiring pseudo-anonymization over simple redaction

FAQ from Mcp Proxy For Pii Pseudo Anonymization

What is pseudo-anonymization and how is it different from redaction?

Pseudo-anonymization replaces real PII with realistic fake data that preserves structure (e.g., [email protected] becomes [email protected]), maintaining semantic meaning for AI analysis. Redaction removes the data entirely, losing context.

What dependencies are required?

For regex mode, no external dependencies. For LLM modes, you need Ollama running locally with a model like llama3.2:3b. Build dependencies require Rust ≥1.70.

Where is the mapping data stored?

Real-to-fake mappings are stored in a SQLite database at the path specified by mapping.database_path in the config. LLM detection cache can also be stored in a separate SQLite database.

What transport does this proxy use?

The proxy communicates via stdio, wrapping the target MCP server’s command. It does not add network transport; it forwards MCP messages between the AI client and the target server.

How are detection results kept consistent across restarts?

A seed set in faker.seed ensures the same real PII always maps to the same fake data. The mapping database persists across restarts, and retention days control how long mappings are kept.