home / mcp / javelin mcp server

Javelin MCP Server

Provides AI guardrails including trust & safety, prompt injection detection, and language policies for MCP-based apps.

Installation

Add the following to your MCP client configuration file.

Configuration

{
  "mcpServers": {
    "abhijitjavelin-javelin-guardrails-mcp-server": {
      "url": "https://your-deployed-url.com/mcp",
      "headers": {
        "MCP_TRANSPORT": "http",
        "JAVELIN_API_KEY": "YOUR_API_KEY"
      }
    }
  }
}

You deploy and run an MCP server that provides guardrails for AI applications. This server integrates with AI security features to detect harmful content, prompt injection attempts, and language usage, enabling you to enforce policies and protect your models in real time.

How to use

Connect your MCP client to the hosted endpoint to leverage the guardrails provided by this server. You can test individual tools to detect prompt injection, assess content for safety categories, and identify the detected language with confidence scores. Use the hosted endpoint to evaluate inputs from your applications and enforce your policies before you allow responses to flow to users.

How to install

Prerequisites: you need Python and a working internet connection to install dependencies and run the server locally.

1) Clone the MCP server repository and install dependencies.

2) Set your API key for authentication with the service.

3) Run the server locally using one of the supplied methods.

Local development and run options

# Method 1: FastMCP CLI using HTTP transport
fastmcp run server.py:mcp --transport http --port 8000

# Method 1 alternative (HTTP via SSE)
fastmcp run server.py:mcp --transport sse --port 8000

# Method 2: Direct execution
export MCP_TRANSPORT=http
export JAVELIN_API_KEY="your-api-key"

python server.py

Notes and testing

Set the transport variable to sse or http depending on your application layer protocol. Test the server locally with the provided test client to ensure all guardrails respond as expected.

Available tools

promptInjectionDetection

Detect prompt injection attempts and jailbreak techniques to prevent model manipulation.

trustSafetyDetection

Analyze content for harmful categories such as violence, weapons, hate speech, crime, sexual content, and profanity.

languageDetection

Detect language with confidence scores and support enforcement of language policies.