# AI Token Cost Optimization — MCP Server

A production-ready [Model Context Protocol (MCP)](https://modelcontextprotocol.io) server that provides **4 tools** to count tokens, estimate API costs, optimize prompts, and compare AI model pricing. Compatible with **Cline**, **Claude Desktop**, and any MCP-compatible client.

---

## Tools Overview

| Tool | Description |
|------|-------------|
| **count_tokens** | Count tokens in any text for GPT-4o, GPT-4o-mini, Claude 3.5 Sonnet, Claude 3.5 Haiku, or Gemini 1.5 Flash. Includes cost estimation. |
| **estimate_cost** | Estimate the full API call cost: input cost + output cost. Supports USD and INR. |
| **optimize_prompt** | Analyze a prompt and get suggestions to reduce token usage. Also recommends the cheapest model for your text. |
| **compare_models** | Compare all 5 models side-by-side for your prompt, sorted from cheapest to most expensive. |

---

## Project Structure

```
my-mcp-server/
├── src/
│   ├── index.ts          # MCP server (4 tools)
│   ├── types.ts          # TypeScript interfaces
│   ├── pricing.ts        # Model pricing data (USD/INR)
│   └── tokenizer.ts      # Token counting (gpt-tokenizer o200k_base)
├── package.json           # Dependencies and scripts
├── tsconfig.json          # TypeScript configuration
├── README.md              # This file
└── DOCUMENTATION.md       # Full process documentation
```

---

## Prerequisites

- [Node.js](https://nodejs.org/) **v18 or higher**
- npm (comes with Node.js)

---

## Quick Start

```bash
# Navigate to the project directory
cd C:\Users\ghosh\Documents\MSS\my-mcp-server

# Install dependencies
npm install

# Run in development mode
npm run dev
```

---

## Usage

### Development Mode (hot-reload with tsx)

```bash
npm run dev
```

### Development Mode (with auto-reload)

```bash
npm run dev:watch
```

### Production Mode

```bash
npm run build    # Compile TypeScript to dist/
npm start        # Run compiled version
```

### TypeScript Type-Check

```bash
npx tsc --noEmit
```

---

## Testing the Server

### Option 1: MCP Inspector (Recommended)

```bash
npx @modelcontextprotocol/inspector npx tsx src/index.ts
```

Opens a web UI at `http://localhost:5173` — browse tools, call them, see responses.

### Option 2: Quick Command-Line Test

Test all 4 tools in one command:

```bash
cd C:\Users\ghosh\Documents\MSS\my-mcp-server

echo '{"jsonrpc":"2.0","id":1,"method":"initialize","params":{"protocolVersion":"2024-11-05","capabilities":{},"clientInfo":{"name":"test","version":"1.0.0"}}}
{"jsonrpc":"2.0","method":"notifications/initialized"}
{"jsonrpc":"2.0","id":2,"method":"tools/call","params":{"name":"count_tokens","arguments":{"text":"Hello World!","model":"gpt-4o"}}}
{"jsonrpc":"2.0","id":3,"method":"tools/call","params":{"name":"compare_models","arguments":{"prompt":"Hello","estimated_output_tokens":100}}}' | npx tsx src/index.ts 2>/dev/null
```

---

## Cline Configuration

Add to your Cline MCP settings:

```json
{
  "mcpServers": {
    "hello-world": {
      "command": "npx",
      "args": [
        "tsx",
        "C:\\Users\\ghosh\\Documents\\MSS\\my-mcp-server\\src\\index.ts"
      ]
    }
  }
}
```

> **macOS/Linux:** Use forward slashes: `"/Users/ghosh/Documents/MSS/my-mcp-server/src/index.ts"`

---

## Tool Details

### count_tokens

**Parameters:**
| Field | Type | Required | Description |
|-------|------|----------|-------------|
| `text` | string | ✅ | The text to count tokens for |
| `model` | string | ✅ | gpt-4o, gpt-4o-mini, claude-3.5-sonnet, claude-3.5-haiku, gemini-1.5-flash |
| `currency` | string | ❌ | USD (default) or INR |

**Example response:**
```json
{
  "model": "gpt-4o",
  "token_count": 7,
  "character_count": 31,
  "tokenizer": "OpenAI o200k_base (via gpt-tokenizer) — exact",
  "estimated_input_cost": "$0.000018",
  "currency": "USD"
}
```

### estimate_cost

**Parameters:**
| Field | Type | Required | Description |
|-------|------|----------|-------------|
| `prompt` | string | ✅ | The input prompt text |
| `estimated_output_tokens` | number | ❌ | Default: 500 |
| `model` | string | ✅ | Model identifier |
| `currency` | string | ❌ | USD (default) or INR |

**Example response:**
```json
{
  "model": "gpt-4o",
  "provider": "OpenAI",
  "input_tokens": 6,
  "output_tokens": 200,
  "input_cost": "$0.000015",
  "output_cost": "$0.002000",
  "total_cost": "$0.002015",
  "currency": "USD"
}
```

### optimize_prompt

**Parameters:**
| Field | Type | Required | Description |
|-------|------|----------|-------------|
| `prompt` | string | ✅ | The prompt to analyze |
| `model` | string | ✅ | Model identifier |

**Example response:**
```json
{
  "original_token_count": 15,
  "character_count": 62,
  "suggested_improvements": [
    { "suggestion": "Replace verbose phrases...", "estimated_savings_percent": 5 },
    { "suggestion": "Consider using GPT-4o-mini...", "estimated_savings_percent": 60 }
  ],
  "best_model_recommendation": "Switch to gpt-4o-mini for ~60% cost savings."
}
```

### compare_models

**Parameters:**
| Field | Type | Required | Description |
|-------|------|----------|-------------|
| `prompt` | string | ✅ | The input prompt text |
| `estimated_output_tokens` | number | ❌ | Default: 500 |
| `currency` | string | ❌ | USD (default) or INR |

**Example response:**
```json
{
  "prompt_character_count": 11,
  "estimated_output_tokens": 100,
  "currency": "USD",
  "comparisons": [
    { "model": "gpt-4o-mini", "total_cost": "$0.000060" },
    { "model": "claude-3.5-haiku", "total_cost": "$0.000503" }
  ],
  "summary": {
    "cheapest": "gpt-4o-mini",
    "most_expensive": "claude-3.5-sonnet"
  }
}
```

---

## Supported Models & Pricing

| Model | Provider | Input (per 1K tokens) | Output (per 1K tokens) |
|-------|----------|----------------------|-----------------------|
| GPT-4o | OpenAI | $0.00250 | $0.01000 |
| GPT-4o-mini | OpenAI | $0.00015 | $0.00060 |
| Claude 3.5 Sonnet | Anthropic | $0.00300 | $0.01500 |
| Claude 3.5 Haiku | Anthropic | $0.00100 | $0.00500 |
| Gemini 1.5 Flash | Google | $0.00150 | $0.00900 |

---

## Technical Details

| Field | Value |
|-------|-------|
| **SDK** | `@modelcontextprotocol/sdk` ^1.29.0 |
| **Token Counting** | `gpt-tokenizer` (o200k_base) for OpenAI; character-based for Claude/Gemini |
| **Transport** | `StdioServerTransport` |
| **Module System** | ES Modules (`"type": "module"`) |
| **Runtime** | Node.js >= 18 |
| **Dev Runner** | `tsx` for TypeScript execution |

---

## Example Cline Prompts

After connecting the server, try asking Cline:

> **"Count the tokens in 'Hello World from my MCP Server!' using GPT-4o"**

> **"Estimate the cost of sending a 100-token prompt with 500 output tokens on Claude 3.5 Sonnet in INR"**

> **"Which is the cheapest model for this prompt: 'Write a poem about artificial intelligence' with 200 output tokens?"**

> **"Analyze this prompt for optimization: 'I would like you to please write a detailed report'"**

---

## License

MIT