# MCP AI Docs

An MCP (Model Context Protocol) server that provides indexed, searchable access to Anthropic Claude and Google Gemini documentation.

## Features

- **Full-text search** across both documentation sources with fuzzy matching
- **Playwright-powered crawling** for JavaScript-rendered pages (Anthropic)
- **TTL-based caching** with memory and file layers
- **4 MCP tools**: search_docs, fetch_page, list_sections, refresh_index

## Installation

```bash
npm install
npx playwright install chromium
npm run build
```

## Usage with Claude Desktop

Add to your Claude Desktop config (`~/.config/claude-desktop/claude_desktop_config.json`):

```json
{
  "mcpServers": {
    "ai-docs": {
      "command": "node",
      "args": ["/path/to/mcp-ai-docs/dist/index.js"]
    }
  }
}
```

## MCP Tools

### search_docs
Search across all indexed documentation.
```json
{
  "query": "streaming",
  "platform": "all",
  "limit": 10,
  "fuzzy": true
}
```

### fetch_page
Fetch full content of a specific documentation page.
```json
{
  "url": "https://ai.google.dev/gemini-api/docs/function-calling",
  "format": "markdown"
}
```

### list_sections
List available documentation sections by platform.
```json
{
  "platform": "all"
}
```

### refresh_index
Rebuild the documentation index.
```json
{
  "platform": "all",
  "force": false
}
```

## Documentation Sources

| Platform | Pages | Method | Speed |
|----------|-------|--------|-------|
| Anthropic | 532 | Playwright (JS-rendered) | ~2.5s/page |
| Gemini | 70 | HTTP (server-rendered) | ~0.6s/page |

## Project Structure

```
src/
├── index.ts              # Entry point
├── server.ts             # MCP server setup
├── config.ts             # Configuration
├── cache/                # TTL caching (memory + file)
├── crawlers/             # Sitemap parsing, page fetching
├── indexer/              # MiniSearch wrapper
├── tools/                # MCP tool implementations
└── utils/                # HTML parsing, rate limiting
```

## License

MIT