# MCP Tool Auditor

> Audit any MCP tool description for LLM reliability. Detect smells, scores quality 0-100, and rewrite for better agent accuracy.

**Zero external APIs. Pure computation. Built for developers who care about agent reliability.**

[![MCP](https://img.shields.io/badge/MCP-compatible-blue)](https://modelcontextprotocol.io)
[![License: MIT](https://img.shields.io/badge/License-MIT-green.svg)](LICENSE)


## Why this exists

Research in 2026 shows that **97% of MCP tool descriptions contain at least one "smell"** (vague purpose, missing usage guidelines, opaque parameters). When descriptions are weak, AI agents pick the wrong tool or waste steps.

This MCP server lets you (or your agent) audit and fix those descriptions before you publish.

## Quick start

```bash
# Install & run with any MCP client
npx -y mcpize connect @princeruhulofficial/tool-auditor --client claude
```

Or add to your Claude / Cursor / other client config:

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

## Tools

| Tool | What it does |
|------|--------------|
| `audit_tool` | Score one tool description (0-100), list smells, give fix suggestions |
| `rewrite_tool_description` | Generate a high-quality rewritten description ready to paste |
| `audit_tool_list` | Audit a full list of tools — detect overcrowding + rank quality |
| `get_best_practices` | Return the current best-practice checklist for MCP tool descriptions |

## Why entrepreneurs care

If you are building AI products or selling MCP servers:

- Agents that pick the wrong tool = wasted tokens + angry users
- High-quality tool descriptions = higher task success rate
- This server is free to run (no API bills) and can be called by your own agents during development

## Pricing (when published on MCPize)

- **Free**: 100 audits / day
- **Pro $9/mo**: 2 000 audits / day

## Development

```bash
npm install
npm run build
npm run test:smoke
```

## License

MIT © Prince Ruhul / Prevalid