ZugaShield

ZugaShield provides an MCP (Multiplayer Cells Protocol) server for safeguarding AI agents by validating inputs, outputs, tool calls, and memory interactions through a layered, zero-dependency security gateway. It enables seamless integration with AI platforms while enforcing threat detection and policy compliance at the MCP boundary, helping you prevent prompt injections, data leakage, and related attacks with sub-15ms overhead.

How to use

You will run the MCP server locally or in your hosting environment and connect your MCP client to it. The server exposes a set of tools that let you scan user messages, LLM outputs, tool calls, memory interactions, and documents before they are processed by or returned from the AI system. You can enable or disable individual defense layers and observe live telemetry through the provided dashboard endpoints when you integrate the FastAPI integration. Your client will send requests to the MCP server’s scan endpoints to validate inputs, tool definitions, and outputs, ensuring security policies are consistently enforced across all interactions.

How to install

Prerequisites: You need Python with pip available on your system. You may also choose to enable optional extras for dashboards and integrations.

pip install zugashield[mcp]

Configuration and usage notes

Configure the MCP server in your client setup to connect to the local or remote MCP runner. The MCP server runs as a separate process that your client can invoke for scanning inputs, tool calls, and memory interactions.

{
  "mcpServers": {
    "zugashield": {
      "command": "zugashield-mcp"
    }
  }
}

Available tools and endpoints

The MCP server exposes a suite of tools for comprehensive security checks. Each tool is designed to scan a specific aspect of the interaction between your AI agent and its environment.

- scan_input: Check user messages for prompt injection
- scan_output: Check LLM responses for data leakage
- scan_tool_call: Validate tool parameters before execution
- scan_tool_definitions: Scan tool schemas for hidden payloads
- scan_memory: Check memory writes for poisoning
- scan_document: Pre-ingestion RAG document scanning
- get_threat_report: Get current threat statistics
- get_config: View active configuration
- update_config: Toggle layers and settings at runtime

Available tools

scan_input

Check user messages for prompt injection to prevent prompt manipulation.

scan_output

Check LLM responses for data leakage and exfiltration.

scan_tool_call

Validate tool parameters and guard against SSRF or harmful commands before tool execution.

scan_tool_definitions

Scan tool schemas for hidden payloads or malicious payload structures.

scan_memory

Check memory writes and reads for poisoning or leakage.

scan_document

Pre-ingestion RAG document scanning to prevent poisoning or leakage.

get_threat_report

Retrieve current threat statistics and threat catalog updates.

get_config

View the active MCP configuration and layer statuses.

update_config

Toggle defense layers and settings at runtime.