MCP Gateway

Overview

MCP Gateway is an advanced intermediary solution for Model Context Protocol (MCP) servers that centralizes and enhances your AI infrastructure.

MCP Gateway acts as an intermediary between LLMs and other MCP servers. It:

1. 📄 Reads server configurations from a mcp.json file located in your root directory.
2. ⚙️ Manages the lifecycle of configured MCP servers.
3. 🛡️ Intercepts requests and responses to sanitize sensitive information.
4. 🔗 Provides a unified interface for discovering and interacting with all proxied MCPs.
5. 🔒 Security Scanner - Analyzes server reputation and security risks before loading MCP servers.

Installation

Python (recommended)

Install the mcp-gateway package:

pip install mcp-gateway

--mcp-json-path - must lead to your mcp.json or claude_desktop_config.json
--plugin or -p - Specify the plugins to enable (can be used multiple times)

Usage

This example enables the basic guardrail for token masking and xetrack tracing plugin for filesystem MCP:

mcp-gateway --mcp-json-path ~/.cursor/mcp.json -p basic -p xetrack

You can add more MCPs that will be under the Gateway by putting the MCP server configuration under the "servers" key.

{
  "mcpServers": {
      "mcp-gateway": {
          "command": "mcp-gateway",
          "args": [
              "--mcp-json-path",
              "~/.cursor/mcp.json",
              "--plugin",
              "basic",
              "--plugin",
              "xetrack"
          ],
          "servers": {
              "filesystem": {
                  "command": "npx",
                  "args": [
                      "-y",
                      "@modelcontextprotocol/server-filesystem",
                      "."
                  ]
              }
          }
      }
  }
}

Get <PYTHON_PATH>

which python

{
  "mcpServers": {
      "mcp-gateway": {
          "command": "<python path>",
          "args": [
            "-m",
            "mcp_gateway.server",
            "--mcp-json-path",
            "<path to claude_desktop_config>",
            "--plugin",
            "basic"
          ],
          "servers": {
              "filesystem": {
                  "command": "npx",
                  "args": [
                      "-y",
                      "@modelcontextprotocol/server-filesystem",
                      "."
                  ]
              }
          }
      }
  }
}

Build the image after clone this repo

docker build -t mcp/gateway .

{
  "mcpServers": {
      "mcp-gateway": {
          "command": "docker",
          "args": [
            "run",
            "--rm",
            "--mount", "type=bind,source=/Users/oro/Projects/playground/mcp-gateway,target=/app",
            "-i",
            "-v", "/Users/oro/.cursor/mcp.json:/config/mcp.json:ro",
            "-e", "LASSO_API_KEY=<LASSO_API_KEY>",
            "-v", "mcp-gateway-logs:/logs",
            "mcp/gateway:latest",
            "--mcp-json-path", "/config/mcp.json",
            "--plugin", "basic",
            "--plugin", "lasso"
          ],
          "servers": {
              "filesystem": {
                  "command": "npx",
                  "args": [
                      "-y",
                      "@modelcontextprotocol/server-filesystem",
                      "."
                  ]
              }
          }
      }
  }
}

In this example we use lasso and basic guardrail to show how we can pass enviroment varabile and arguments to the docker and how we can mount storage for the filesystem MCP. The Docker image can be built with optional dependencies required by certain plugins (e.g., presidio).
Use the INSTALL_EXTRAS build argument during the docker build command. Provide a comma-separated string of the desired extras: "presidio,xetrack"

Quickstart

Masking Sensitive Information

MCP Gateway will automatically mask the sensitive token in the response, preventing exposure of credentials while still providing the needed functionality.

1. Create a file with sensitive information:

   echo 'HF_TOKEN = "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"' > tokens.txt
   

2. When an agent requests to read this file through MCP Gateway:

   • Recommend to test with sonnet 3.7

   Use your mcp-gateway tools to read the ${pwd}/tokens.txt and return the HF_TOKEN
   

Output:

Usage

Start the MCP Gateway server with python_env config on this repository root:

mcp-gateway -p basic -p presidio

You can also debug the server using:

LOGLEVEL=DEBUG mcp-gateway --mcp-json-path ~/.cursor/mcp.json -p basic -p presidio

Tools

Here are the tools the MCP is using to create a proxy to the other MCP servers

• get_metadata - Provides information about all available proxied MCPs to help LLMs choose appropriate tools and resources
• run_tool - Executes capabilities from any proxied MCP after sanitizing the request and response

Plugins

Contribute

For more details on how the plugin system works, how to create your own plugins, or how to contribute, please see the Plugin System Documentation.

Guardrails

MCP Gateway supports various plugins to enhance security and functionality. Here's a summary of the built-in guardrail plugins:

| Name | PII Masking | Token/Secret Masking | Custom Policy | Prompt Injection | Harmful Content | | :---------- | :----------------------------------------------------------------------- | :----------------------------------------------------------------------------------- | :-----------: | :------------------: | :-------------: | | basic | ❌ | ✅ | ❌ | ❌ | ❌ | | presidio | ✅ | ❌ | ❌ | ❌ | ❌ | | lasso | ✅ | ✅ | ✅ | ✅ | ✅ |

Note: To use the presidio plugin, you need to install it separately: pip install mcp-gateway[presidio].

Basic

mcp-gateway -p basic

Masking basic secerts

• azure client secret
• github tokens
• github oauth
• gcp api key
• aws access token
• jwt token
• gitlab session cookie
• huggingface access token
• microsoft teams webhook
• slack app token

Presidio

mcp-gateway -p presidio

Presidio is identification and anonymization package

• Credit Card
• IP
• Email
• Phone
• SSN
• Etc

Lasso

mcp-gateway -p lasso

Prerequisites

• Obtain a Lasso API key by signing up at Lasso Security.

To use Lasso Security's advanced AI safety guardrails, update your mcp.json configuration as follows:

1. Add the LASSO_API_KEY=<YOUR-API-KEY> to your environment variable or in the "env" section.
2. Insert other MCP servers configuration under key servers

Example:

{
  "mcpServers": {
      "mcp-gateway": {
          "command": "mcp-gateway",
          "args": [
              "--mcp-json-path",
              "~/.cursor/mcp.json",
              "-p",
              "lasso"
          ],
          "env": {
              "LASSO_API_KEY": "<lasso_token>"
          },
          "servers": {
              "filesystem": {
                  "command": "npx",
                  "args": [
                      "-y",
                      "@modelcontextprotocol/server-filesystem",
                      "."
                  ]
              }
          }
      }
  }
}

Features

🔍 Full visibility into MCP interactions with an Always-on monitoring.

🛡️ Mitigate GenAI-specific threats like prompt injection and sensitive data leakage in real-time with built-in protection that prioritizes security from deployment.

✨ Use flexible, natural language to craft security policies tailored to your business's unique needs.

⚡ Fast and easy installation for any deployment style. Monitor data flow to and from MCP in minutes with an intuitive, user-friendly dashboard.

The Lasso guardrail checks content through Lasso's API for security violations before processing requests and responses.

Read more on our website 👉 Lasso Security.

Tracing

Xetrack

xetrack is a lightweight package to track ml experiments, benchmarks, and monitor stractured data.

We can use it to debug and monitor tool calls with logs (loguru) or [duckd