# AgentCost — AI Agent Spend Analysis

MCP server that analyzes AI agent session logs to find token waste and optimization opportunities.

**Free preview tier** — see where your money goes before paying for detailed analysis.

## Quick Start

Add to your Claude Desktop or MCP client config:

```json
{
  "mcpServers": {
    "agentcost": {
      "command": "npx",
      "args": ["-y", "alphaforensics-mcp"]
    }
  }
}
```

No API key needed for the free preview tier.

## Tools

### `agentcost_preview` (Free)

Analyze your Claude Code session logs at no cost. Returns:
- Total cost breakdown by model
- Token counts (input/output/cache)
- Tool usage frequency
- Top 5 most expensive turns

```
Upload your JSONL from ~/.claude/projects/ to see where your money goes.
```

### `agentcost_analyze` ($5-$15 via x402)

Full analysis with actionable recommendations:
- **Pro ($5)**: Waste pattern detection, per-tool cost attribution, optimization recommendations
- **Full ($15)**: + industry benchmarks, savings projections, exportable report

### `agentcost_sample` (Free)

Get a sample report to see the output format before submitting real data.

## How It Works

1. Find your Claude Code session logs in `~/.claude/projects/`
2. Use `agentcost_preview` to get a free cost breakdown
3. If you want waste detection and recommendations, use `agentcost_analyze`

Paid analysis uses [x402 micropayments](https://the402.ai) on Base mainnet — no API keys or subscriptions needed.

## API Endpoints

| Endpoint | Price | Description |
|----------|-------|-------------|
| `POST /forensics/preview` | Free | Cost breakdown + top-5 hot turns |
| `POST /forensics/analyze` | $5-$15 | Waste patterns + recommendations |
| `GET /forensics/sample` | Free | Sample report |

Base URL: `https://api.agentsconsultants.com`

## License

MIT