AI Testing MCP

Model Context Protocol (MCP) server for comprehensive AI testing, evaluation, and quality assurance.

Overview

AI Testing MCP provides standardized testing methodologies, evaluation metrics, and automated testing workflows for AI/ML systems. It implements the Model Context Protocol for seamless integration with AI development tools.

Features

Testing Capabilities

• Unit Tests: Component-level testing
• Integration Tests: End-to-end workflow testing
• Performance Tests: Latency, throughput, resource usage
• Security Tests: Adversarial attacks, prompt injection
• Quality Tests: Output validation, consistency checks

Evaluation Metrics

• Accuracy: Precision, recall, F1 score
• Quality: Coherence, relevance, fluency
• Safety: Toxicity, bias detection
• Performance: Response time, token usage
• Cost: API costs, compute resources

MCP Integration

• Standard Protocol: Implements MCP specification
• Tool Definitions: Testing tools for AI assistants
• Resource Management: Test data and configuration
• Prompt Templates: Reusable test prompts

Quick Start

# Clone the repository
git clone https://github.com/groovy-web/ai-testing-mcp.git
cd ai-testing-mcp

# Install dependencies
npm install

# Configure environment
cp .env.example .env
# Edit with your API keys

# Start the MCP server
npm start

# Use with MCP client
# Add server configuration to your MCP client

MCP Server Configuration

{
  "mcpServers": {
    "ai-testing": {
      "command": "node",
      "args": ["/path/to/ai-testing-mcp/dist/index.js"],
      "env": {
        "OPENAI_API_KEY": "your-key",
        "ANTHROPIC_API_KEY": "your-key"
      }
    }
  }
}

Available Tools

Testing Tools

run_test_suite

Execute a comprehensive test suite for an AI model.

{
  "name": "run_test_suite",
  "description": "Run tests on an AI model",
  "inputSchema": {
    "type": "object",
    "properties": {
      "model": { "type": "string" },
      "testCategory": {
        "type": "string",
        "enum": ["accuracy", "performance", "security", "quality"]
      },
      "testCases": { "type": "array" }
    }
  }
}

evaluate_output

Evaluate AI model outputs against metrics.

{
  "name": "evaluate_output",
  "description": "Evaluate model output quality",
  "inputSchema": {
    "type": "object",
    "properties": {
      "output": { "type": "string" },
      "expected": { "type": "string" },
      "metrics": { "type": "array" }
    }
  }
}

generate_test_cases

Generate test cases for specific scenarios.

{
  "name": "generate_test_cases",
  "description": "Generate test cases",
  "inputSchema": {
    "type": "object",
    "properties": {
      "scenario": { "type": "string" },
      "count": { "type": "number" }
    }
  }
}

Repository Structure

ai-testing-mcp/
├── docs/                    # Documentation
│   ├── mcp-protocol.md
│   ├── testing-guide.md
│   └── metrics.md
├── examples/                # Usage examples
│   ├── basic-testing/
│   ├── custom-metrics/
│   └── integration-examples/
├── src/                     # Source code
│   ├── server/              # MCP server
│   ├── tools/               # Tool implementations
│   ├── metrics/             # Evaluation metrics
│   └── tests/               # Test definitions
└── schemas/                 # JSON schemas

Usage Examples

Basic Testing

from mcp_client import MCPClient

client = MCPClient("ai-testing")

# Run accuracy tests
result = client.call_tool("run_test_suite", {
    "model": "gpt-4",
    "testCategory": "accuracy",
    "testCases": [
        {
            "input": "What is 2+2?",
            "expected": "4"
        }
    ]
})

print(f"Accuracy: {result.metrics.accuracy}")

Custom Metrics

# Evaluate with custom metrics
evaluation = client.call_tool("evaluate_output", {
    "output": model_response,
    "expected": expected_response,
    "metrics": [
        "exact_match",
        "semantic_similarity",
        "coherence",
        "relevance"
    ]
})

Documentation

• MCP Protocol - Protocol specification
• Testing Guide - How to write tests
• Metrics Reference - Available evaluation metrics
• API Documentation - Complete API reference

Test Categories

1. Accuracy Tests

• Exact match
• Semantic similarity
• Factual correctness
• Mathematical accuracy

2. Performance Tests

• Response time
• Throughput
• Token efficiency
• Resource usage

3. Security Tests

• Prompt injection
• Jailbreak attempts
• Toxic content
• Bias detection

4. Quality Tests

• Coherence
• Fluency
• Relevance
• Completeness

Configuration

module.exports = {
  // Models to test
  models: [
    {
      name: "gpt-4",
      provider: "openai",
      apikey: process.env.OPENAI_API_KEY
    },
    {
      name: "claude-3-opus",
      provider: "anthropic",
      apikey: process.env.ANTHROPIC_API_KEY
    }
  ],

  // Test configurations
  tests: {
    accuracy: {
      enabled: true,
      threshold: 0.95
    },
    performance: {
      maxLatency: 2000,
      maxTokens: 1000
    },
    security: {
      enabled: true,
      strict: true
    }
  },

  // Output format
  reports: {
    format: "json",
    destination: "./test-results"
  }
};

Running Tests

# Run all tests
npm test

# Run specific category
npm test -- --category accuracy

# Run with specific model
npm test -- --model gpt-4

# Generate report
npm run report

Contributing

We welcome contributions! Please see CONTRIBUTING.md for guidelines.

License

This project is licensed under the MIT License - see LICENSE for details.

Code of Conduct

Please read CODE_OF_CONDUCT.md to understand our community standards.

Support

• GitHub Issues: Bug reports and feature requests
• Discussions: Community questions
• MCP Documentation: https://modelcontextprotocol.io

Related Projects

• MCP SDK
• Claude Code
• AI Testing Benchmarks

Related Repositories

Explore more open-source tools from Groovy Web:

• langchain-multi-agent-example -- Multi-agent systems tutorial with LangChain
• rag-system-pgvector -- Production RAG with PostgreSQL + pgvector
• rag-systems-production -- Enterprise-grade RAG systems
• ai-testing-mcp -- AI testing via Model Context Protocol
• edge-computing-starter -- Cloudflare Workers + Hono template
• claude-code-workflows -- Workflows for Claude Code
• groovy-web-ai-agents -- Production AI agent configs
• groovy-web-examples -- Groovy/Grails examples