# Agent Reliability MCP Server

**Compute AI agent reliability scores, success rates, latency stats and failure patterns — pure math, zero API cost.**

This MCP server gives any AI agent (Claude, Cursor, ChatGPT, etc.) the ability to analyze how reliable other AI agents are using only the numbers you feed it. Perfect for entrepreneurs building agent products who want quick, trustworthy metrics without expensive observability platforms.

## Why this exists

When you ship an AI agent, it sometimes fails. Counting successes by hand is boring. This server does the hard math for you in one tool call.

## Tools

| Tool | What it does |
|------|--------------|
| `score_agent_reliability` | Gives an overall 0-100 reliability score + letter grade |
| `calculate_success_rate` | Success % with statistical confidence interval |
| `analyze_latency` | Mean, median, p95, p99 latency numbers |
| `detect_failure_patterns` | Finds the most common error messages |
| `simulate_reliability` | Monte-Carlo projection of future success |
| `compare_agents` | Which of two agents is more reliable? |

## Quick start (after publish)

```bash
npx @mcpize/cli install agent-reliability-mcp
```

Or add to your MCP client config.

## For entrepreneurs

- Zero running cost (no external APIs)
- Helps you decide which agent version to ship
- Works offline
- Ready for MCPize marketplace

## License

MIT