MCP Presidio

MCP Presidio is an MCP server that detects and anonymizes PII in text and structured data using Microsoft Presidio. It enables safe handling of sensitive information by providing robust PII recognition and flexible anonymization options for both plain text and JSON-like data structures.

How to use

You run the MCP Presidio server locally or in a container and connect your MCP client to it through stdio. Use the server to detect PII in text or JSON-like data and then apply anonymization accordingly. You can analyze single texts, anonymize them, or process batches. You can also extend the recognizers with custom patterns and test detection with validation tools.

How to install

Prerequisites you need before installation:

• Python 3.10 or higher

• Docker 20.10 or higher

Choose your installation method and follow the steps below.

Docker installation (recommended for production)

Quick start with Docker to run the MCP Presidio server in a self-contained container.

# Clone the repository
git clone https://github.com/cmalpass/mcp-presidio.git
cd mcp-presidio

# Build the Docker image
docker build -t mcp-presidio .

# Run the container with stdio (default)
docker run -i mcp-presidio

Using Docker Compose

You can build and start the container with Docker Compose, then view logs or stop the service as needed.

# Clone the repository
git clone https://github.com/cmalpass/mcp-presidio.git
cd mcp-presidio

# Build and start the container
docker-compose up -d

# View logs
docker-compose logs -f

# Stop the container
docker-compose down

Configuring Claude Desktop with Docker

If you want Claude Desktop to run the Presidio MCP server in a Docker container, configure the MCP server entry to launch the container.

{
  "mcpServers": {
    "presidio": {
      "command": "docker",
      "args": [
        "run",
        "-i",
        "--rm",
        "mcp-presidio:latest"
      ],
      "env": {}
    }
  }
}

Docker image details

The Docker image includes Python 3.11 slim, the Presidio components, spaCy, and an English language model. It runs with a non-root user and uses a multi-stage build to keep the image small.

Advanced Docker usage

Interact with the container for debugging, add more language models, or mount a local configuration.

docker run -it mcp-presidio bash

Python installation (quick install)

If you prefer a Python-based setup, run the interactive installation script or install manually.

# Clone the repository
git clone https://github.com/cmalpass/mcp-presidio.git
cd mcp-presidio

# Run the installation script
./install.sh
# or
python install.py

Python installation (manual)

For a manual setup, install the package and download the English spaCy model.

# Clone the repository
git clone https://github.com/cmalpass/mcp-presidio.git
cd mcp-presidio

# Install the package
pip install -e .

# Download required spaCy language model (for English)
python -m spacy download en_core_web_lg

Running the server

Start the server using stdio transport, suitable for MCP clients.

mcp-presidio

or

python -m mcp_presidio.server

Configuring with Claude Desktop

Add the MCP server configuration to Claude Desktop so it can launch the server as part of your workflow.

{
  "mcpServers": {
    "presidio": {
      "command": "python",
      "args": ["-m", "mcp_presidio.server"],
      "env": {}
    }
  }
}

Example usage in conversations

The server powers tools that detect and anonymize PII in text. You can run a detection or an anonymization flow through an MCP client and inspect the results.

Supported PII entity types

The server supports 25+ PII types, including personal, contact, financial, government IDs, international IDs, location, medical, and other identifiers. Use the get_supported_entities tool to list all available types for your language.

Anonymization operators

Choose how you replace or remove PII with operators like replace, redact, hash, mask, encrypt, or keep.

Additional features

Custom recognizers enable domain-specific patterns, batch processing handles multiple documents efficiently, and the server architecture integrates MCP FastMCP with Presidio and spaCy for fast, accurate detection.

Security considerations

All processing happens locally, and the server communicates via stdio with MCP clients. Docker deployment provides isolation, and the container can run as a non-root user for enhanced security.

Development and tests

To run development tests locally, install dev dependencies, then run the test suite with pytest.

pip install -e ".[dev]"
pytest tests/

Project structure and license

The project includes the MCP server implementation under src/mcp_presidio/server.py, along with tests, a Docker setup, and build configurations. The license is MIT.

Notes and caveats

Be mindful of privacy considerations when analyzing text with external agents. If sensitive data will be sent to third-party services, consider local LLMs or private deployments to keep data in your environment.

Available tools

analyze_text

Detect PII entities in text with confidence scores.

anonymize_text

Anonymize detected PII using various operators such as replace, redact, hash, mask, encrypt, or keep.

get_supported_entities

List all supported PII entity types for your language.

add_custom_recognizer

Add custom PII recognition patterns to extend detection capabilities.

batch_analyze

Analyze multiple texts for PII in a single call.

batch_anonymize

Anonymize PII across multiple texts in one operation.

get_anonymization_operators

List available methods for anonymizing detected PII.

analyze_structured_data

Detect PII in JSON or structured data.

anonymize_structured_data

Anonymize PII found in structured data.

validate_detection

Validate detection accuracy with test metrics.