llms-full.txt

# markgrab-plugin
> Claude Code plugin for universal web content extraction — any URL to LLM-ready markdown

Wraps [markgrab](https://github.com/QuartzUnit/markgrab) as Claude Code skills + MCP server.
Install: `pip install markgrab` (core) | `pip install "markgrab[all]"` (all extractors)

## Skills

### `extract-url`
Extract content from a web URL and convert to LLM-ready markdown.

**Supported content types** (auto-detected from URL):
- HTML pages — content density filtering, auto-fallback to Playwright for JS-heavy sites
- YouTube — transcript extraction with timestamps, multi-language
- PDF — text extraction with page structure
- DOCX — paragraph and heading extraction

**Command:**
```bash
python -m markgrab <URL> [OPTIONS]
```
| option | description |
|--------|-------------|
| `--format` | Output: `markdown` (default), `text`, `json` |
| `--browser` | Force Playwright rendering |
| `--max-chars N` | Limit output length (default: 30000) |
| `--stealth` | Anti-bot stealth scripts (browser only) |

### `extract-file`
Convert local PDF or DOCX files to clean markdown.

**Supported file types:**
- PDF — requires `markgrab[pdf]`
- DOCX — requires `markgrab[docx]`

**Command:**
```bash
python -m markgrab <FILE_PATH> [OPTIONS]
```
| option | description |
|--------|-------------|
| `--format` | Output: `markdown` (default), `text`, `json` |
| `--max-chars N` | Limit output length (default: 50000) |

### `batch-extract`
Extract content from multiple URLs at once.

**Python API:**
```python
import asyncio
from markgrab import extract

async def batch_extract(urls: list[str]) -> list:
    results = []
    for url in urls:
        try:
            result = await extract(url, max_chars=30_000)
            results.append({
                "url": url,
                "title": result.title,
                "markdown": result.markdown,
                "word_count": result.word_count
            })
        except Exception as e:
            results.append({"url": url, "error": str(e)})
    return results
```

## MCP Server

Configuration (`.mcp.json`):
```json
{
  "mcpServers": {
    "markgrab": {
      "command": "markgrab-mcp",
      "args": []
    }
  }
}
```

**MCP Tools:**
- `extract_url` — Extract single URL to markdown
- `extract_multiple` — Batch extract multiple URLs

## Plugin Configuration

`.claude-plugin/plugin.json`:
```json
{
  "name": "markgrab",
  "version": "0.1.1",
  "author": "QuartzUnit",
  "license": "MIT",
  "repository": "https://github.com/QuartzUnit/markgrab"
}
```

## markgrab Python API Reference

```python
async def extract(url: str, *, engine=None, max_chars: int = 50_000,
                  use_browser: bool = False, stealth: bool = False,
                  timeout: float = 30.0, proxy: str | None = None,
                  locale: str | None = None,
                  browser_fallback: bool = True) -> ExtractResult
```

```python
@dataclass
class ExtractResult:
    title: str
    text: str
    markdown: str
    word_count: int
    language: str | None
    url: str
    extracted_at: str
    engine_used: str
    metadata: dict
```