AI Cost Optimizer

Intelligent LLM router that reduces AI API costs by up to 60% without sacrificing quality.

What It Does

Automatically routes your prompts to the most cost-efficient model for each task:

• Smart routing analyzes each prompt and selects the optimal provider
• Cost tracking monitors spending across all requests
• Multi-provider support works with Google Gemini, Anthropic Claude, OpenRouter, and more
• MCP integration seamlessly integrates with Claude Desktop

No configuration needed - just add your API keys and let the optimizer do the rest.

Quick Start

1. Get API Keys

You need at least one of these:

• Google Gemini (recommended - free tier): https://aistudio.google.com/app/apikey
• Anthropic Claude: https://console.anthropic.com/
• OpenRouter (all models): https://openrouter.ai/keys

2. Setup

# Clone or navigate to project
cd ai-cost-optimizer

# Copy environment template
cp .env.example .env

# Edit .env and add your API key(s)
nano .env

# Install dependencies
pip install -r requirements.txt

3. Start the Service

# Run the optimizer
python app/main.py

# You should see:
# "AI Cost Optimizer initialized with providers: ['gemini']"
# "Uvicorn running on http://0.0.0.0:8000"

Keep this terminal running!

4. Configure Claude Desktop (Optional)

Edit your Claude Desktop config:

macOS: ~/Library/Application Support/Claude/claude_desktop_config.json

Add the MCP server:

{
  "mcpServers": {
    "ai-cost-optimizer": {
      "command": "python3",
      "args": ["/ABSOLUTE/PATH/TO/ai-cost-optimizer/mcp/server.py"],
      "env": {
        "COST_OPTIMIZER_API_URL": "http://localhost:8000"
      }
    }
  }
}

Important: Use the absolute path to mcp/server.py on your system!

5. Test It!

Via API

curl -X POST http://localhost:8000/complete \
  -H "Content-Type: application/json" \
  -d '{"prompt": "What is quantum computing?", "max_tokens": 1000}'

Response:

{
  "response": "Quantum computing is...",
  "provider": "gemini",
  "model": "gemini-1.5-flash",
  "tokens_in": 4,
  "tokens_out": 50,
  "cost": 0.000015,
  "total_cost_today": 0.000015
}

Via Claude Desktop

Please use the cost optimizer to answer: What is quantum computing?

You should see the response along with cost tracking.

API Endpoints

Complete a Prompt

POST /complete
{
  "prompt": "Your prompt here",
  "max_tokens": 1000
}

Returns the response with cost breakdown.

Get Usage Statistics

GET /stats

Returns total cost and request statistics.

Get Provider Status

GET /providers

Lists all available providers and their status.

Health Check

GET /health

Returns service health status.

Environment Variables

Create a .env file with your API keys:

# Provider API keys (add at least one)
GOOGLE_API_KEY=your-key-here
ANTHROPIC_API_KEY=your-key-here
OPENROUTER_API_KEY=your-key-here
CEREBRAS_API_KEY=your-key-here

# Optional configuration
PORT=8000
LOG_LEVEL=INFO

How It Works

The optimizer uses an intelligent routing engine that:

1. Analyzes incoming prompts
2. Selects the most cost-efficient provider for the task
3. Tracks all costs and usage metrics
4. Automatically falls back to alternative providers if needed

Result: Up to 60% cost reduction compared to always using premium models.

Troubleshooting

Service won't start

# Check if port 8000 is in use
lsof -i :8000

# Verify API keys are set
cat .env

MCP tool not appearing in Claude Desktop

1. Verify absolute path in claude_desktop_config.json
2. Check service is running: curl http://localhost:8000/health
3. Completely quit and restart Claude Desktop (Cmd+Q on Mac)
4. Check Claude Desktop logs

API errors

• Verify API keys are valid and active
• Check provider service status
• Review logs for detailed error messages

Testing

# Install test dependencies
pip install pytest pytest-asyncio

# Run test suite
pytest

# Run with coverage
pytest --cov=app tests/

License

MIT - do whatever you want with it!

Questions?

This is a learning project built to help developers reduce AI costs. Feel free to fork, modify, and make it yours!