Shield MCP

A security middleware for Model Context Protocol (MCP) servers that enhances security and monitoring capabilities without modifying the official SDK. This package provides tools for securing and monitoring MCP tool calls, following the best practices outlined in the MCP documentation. Abstract yourself while interact at MCP development.

Table of Contents

• Features
• Requirements
• Installation
• Quick Start
• Components
• Best Practices
• Development
• Roadmap
• Acknowledgments

Features

• Tool Access Control: Whitelist-based access control for MCP tools
• Result Sanitization: Configurable sanitization of tool outputs
• Structured Logging: Comprehensive audit logging using structlog
• Rate Limiting: Token bucket algorithm for rate limiting
• Error Handling: Standardized error handling and formatting
• MCP Inspector Compatible: Works seamlessly with the MCP Inspector tool

Requirements

System Requirements

• Python 3.8 or higher
• pip (Python package installer)
• virtualenv (recommended for development)

Quick Start

from shieldmcp import secure_tool
from shieldmcp.sanitizers import ToolSanitizer
from shieldmcp.rate_limit import RateLimitConfig

# Define allowed tools
ALLOWED_TOOLS = {"search", "read_file", "write_file"}

# Create a text sanitizer
text_sanitizer = ToolSanitizer.createTextSanitizer(
    max_length=1000,
    sensitive_patterns=[
        r"\b\d{16}\b",  # Credit card numbers
        r"\b[A-Za-z0-9._%+-]+@[A-Za-z0-9.-]+\.[A-Z|a-z]{2,}\b"  # Email addresses
    ]
)

# Configure rate limiting
rate_limit = RateLimitConfig(
    requests_per_minute=60,  # 1 request per second
    burst_size=10  # Allow bursts of up to 10 requests
)

# Apply the decorator to your MCP tools
@secure_tool(
    allowed_tools=ALLOWED_TOOLS,
    sanitize_fn=text_sanitizer,
    user_id="user123",
    session_id="session456",
    rate_limit=rate_limit
)
def search(query: str):
    # Your tool implementation
    return results

Components

Decorators (decorators.py)

The main @secure_tool decorator that orchestrates all security features:

@secure_tool(
    allowed_tools={"tool1", "tool2"},  # Set of allowed tool names
    sanitize_fn=your_sanitizer,        # Optional result sanitization function
    user_id="user123",                 # Optional user identifier
    session_id="session456",           # Optional session identifier
    rate_limit=RateLimitConfig(        # Optional rate limit configuration
        requests_per_minute=60,
        burst_size=10
    )
)
def your_tool():
    pass

Audit Logging (audit.py)

Structured logging using structlog:

from shieldmcp import ToolAudit

audit = ToolAudit()
audit.logToolCallStart(
    tool_name="search",
    args={"query": "test"},
    user_id="user123"
)

Access Control (access.py)

Tool access validation:

from shieldmcp import ToolAccess

access = ToolAccess(allowed_tools={"tool1", "tool2"})
access.validateToolAccess("tool1")  # Raises ValueError if not allowed

Sanitizers (sanitizers.py)

Result sanitization utilities:

from shieldmcp import ToolSanitizer

# Create a custom sanitizer
sanitizer = ToolSanitizer.createTextSanitizer(
    max_length=1000,
    sensitive_patterns=[r"\b\d{16}\b"]
)

# Use it directly
clean_text = sanitizer("Your text with sensitive data")

Rate Limiting (rate_limit.py)

Token bucket rate limiting:

from shieldmcp import RateLimitConfig

# Configure rate limits
config = RateLimitConfig(
    requests_per_minute=60,
    burst_size=10
)

Best Practices

Tool Access Control

• Always define a whitelist of allowed tools
• Use the most restrictive set of tools possible
• Regularly review and update the whitelist

Result Sanitization

• Sanitize all text output
• Define patterns for sensitive data
• Set reasonable length limits

Logging

• Include user and session IDs when available
• Log both successful and failed operations
• Use structured logging for better analysis

Rate Limiting

• Set appropriate limits based on tool complexity
• Consider burst sizes for better user experience
• Monitor rate limit hits in logs

Development

Setup Development Environment

# Clone the repository
git clone https://github.com/shieldmcp/shieldmcp.git
cd modelShield

# Create virtual environment
python -m venv venv
source venv/bin/activate  # or `venv\Scripts\activate` on Windows

# Install development dependencies
pip install -r requirements.txt

Running Tests

pytest tests/

Roadmap

Planned Features

• Support for Clerk MCP and Github MCP
• Extended documentation
• TypeScript support

Acknowledgments

• Model Context Protocol for the protocol specification
• structlog for structured logging

Feel free to make any inquiries.