kreuzberg 3.6.1__py3-none-any.whl → 3.7.0__py3-none-any.whl

kreuzberg/_mcp/__init__.py

@@ -0,0 +1,5 @@
+"""MCP server for Kreuzberg text extraction."""
+
+from .server import mcp
+
+__all__ = ["mcp"]

kreuzberg/_mcp/server.py

@@ -0,0 +1,227 @@
+"""Kreuzberg MCP server implementation."""
+
+from __future__ import annotations
+
+import base64
+from typing import Any
+
+from mcp.server import FastMCP
+from mcp.types import TextContent
+
+from kreuzberg._types import ExtractionConfig, OcrBackendType
+from kreuzberg.extraction import extract_bytes_sync, extract_file_sync
+
+# Create the MCP server
+mcp = FastMCP("Kreuzberg Text Extraction")
+
+
+@mcp.tool()
+def extract_document(  # noqa: PLR0913
+    file_path: str,
+    mime_type: str | None = None,
+    force_ocr: bool = False,
+    chunk_content: bool = False,
+    extract_tables: bool = False,
+    extract_entities: bool = False,
+    extract_keywords: bool = False,
+    ocr_backend: OcrBackendType = "tesseract",
+    max_chars: int = 1000,
+    max_overlap: int = 200,
+    keyword_count: int = 10,
+    auto_detect_language: bool = False,
+) -> dict[str, Any]:
+    """Extract text content from a document file.
+
+    Args:
+        file_path: Path to the document file
+        mime_type: MIME type of the document (auto-detected if not provided)
+        force_ocr: Force OCR even for text-based documents
+        chunk_content: Split content into chunks
+        extract_tables: Extract tables from the document
+        extract_entities: Extract named entities
+        extract_keywords: Extract keywords
+        ocr_backend: OCR backend to use (tesseract, easyocr, paddleocr)
+        max_chars: Maximum characters per chunk
+        max_overlap: Character overlap between chunks
+        keyword_count: Number of keywords to extract
+        auto_detect_language: Auto-detect document language
+
+    Returns:
+        Extracted content with metadata, tables, chunks, entities, and keywords
+    """
+    config = ExtractionConfig(
+        force_ocr=force_ocr,
+        chunk_content=chunk_content,
+        extract_tables=extract_tables,
+        extract_entities=extract_entities,
+        extract_keywords=extract_keywords,
+        ocr_backend=ocr_backend,
+        max_chars=max_chars,
+        max_overlap=max_overlap,
+        keyword_count=keyword_count,
+        auto_detect_language=auto_detect_language,
+    )
+
+    result = extract_file_sync(file_path, mime_type, config)
+    return result.to_dict()
+
+
+@mcp.tool()
+def extract_bytes(  # noqa: PLR0913
+    content_base64: str,
+    mime_type: str,
+    force_ocr: bool = False,
+    chunk_content: bool = False,
+    extract_tables: bool = False,
+    extract_entities: bool = False,
+    extract_keywords: bool = False,
+    ocr_backend: OcrBackendType = "tesseract",
+    max_chars: int = 1000,
+    max_overlap: int = 200,
+    keyword_count: int = 10,
+    auto_detect_language: bool = False,
+) -> dict[str, Any]:
+    """Extract text content from document bytes.
+
+    Args:
+        content_base64: Base64-encoded document content
+        mime_type: MIME type of the document
+        force_ocr: Force OCR even for text-based documents
+        chunk_content: Split content into chunks
+        extract_tables: Extract tables from the document
+        extract_entities: Extract named entities
+        extract_keywords: Extract keywords
+        ocr_backend: OCR backend to use (tesseract, easyocr, paddleocr)
+        max_chars: Maximum characters per chunk
+        max_overlap: Character overlap between chunks
+        keyword_count: Number of keywords to extract
+        auto_detect_language: Auto-detect document language
+
+    Returns:
+        Extracted content with metadata, tables, chunks, entities, and keywords
+    """
+    content_bytes = base64.b64decode(content_base64)
+
+    config = ExtractionConfig(
+        force_ocr=force_ocr,
+        chunk_content=chunk_content,
+        extract_tables=extract_tables,
+        extract_entities=extract_entities,
+        extract_keywords=extract_keywords,
+        ocr_backend=ocr_backend,
+        max_chars=max_chars,
+        max_overlap=max_overlap,
+        keyword_count=keyword_count,
+        auto_detect_language=auto_detect_language,
+    )
+
+    result = extract_bytes_sync(content_bytes, mime_type, config)
+    return result.to_dict()
+
+
+@mcp.tool()
+def extract_simple(
+    file_path: str,
+    mime_type: str | None = None,
+) -> str:
+    """Simple text extraction from a document file.
+
+    Args:
+        file_path: Path to the document file
+        mime_type: MIME type of the document (auto-detected if not provided)
+
+    Returns:
+        Extracted text content as a string
+    """
+    config = ExtractionConfig()
+    result = extract_file_sync(file_path, mime_type, config)
+    return result.content
+
+
+@mcp.resource("config://default")
+def get_default_config() -> str:
+    """Get the default extraction configuration."""
+    config = ExtractionConfig()
+    return str(config.__dict__)
+
+
+@mcp.resource("config://available-backends")
+def get_available_backends() -> str:
+    """Get available OCR backends."""
+    return "tesseract, easyocr, paddleocr"
+
+
+@mcp.resource("extractors://supported-formats")
+def get_supported_formats() -> str:
+    """Get supported document formats."""
+    return """
+    Supported formats:
+    - PDF documents
+    - Images (PNG, JPG, JPEG, TIFF, BMP, WEBP)
+    - Office documents (DOCX, PPTX, XLSX)
+    - HTML files
+    - Text files (TXT, CSV, TSV)
+    - And more...
+    """
+
+
+@mcp.prompt()
+def extract_and_summarize(file_path: str) -> list[TextContent]:
+    """Extract text from a document and provide a summary prompt.
+
+    Args:
+        file_path: Path to the document file
+
+    Returns:
+        Extracted content with summarization prompt
+    """
+    result = extract_file_sync(file_path, None, ExtractionConfig())
+
+    return [
+        TextContent(
+            type="text",
+            text=f"Document Content:\n{result.content}\n\nPlease provide a concise summary of this document.",
+        )
+    ]
+
+
+@mcp.prompt()
+def extract_structured(file_path: str) -> list[TextContent]:
+    """Extract text with structured analysis prompt.
+
+    Args:
+        file_path: Path to the document file
+
+    Returns:
+        Extracted content with structured analysis prompt
+    """
+    config = ExtractionConfig(
+        extract_entities=True,
+        extract_keywords=True,
+        extract_tables=True,
+    )
+    result = extract_file_sync(file_path, None, config)
+
+    content = f"Document Content:\n{result.content}\n\n"
+
+    if result.entities:
+        content += f"Entities: {[f'{e.text} ({e.type})' for e in result.entities]}\n\n"
+
+    if result.keywords:
+        content += f"Keywords: {[f'{kw[0]} ({kw[1]:.2f})' for kw in result.keywords]}\n\n"
+
+    if result.tables:
+        content += f"Tables found: {len(result.tables)}\n\n"
+
+    content += "Please analyze this document and provide structured insights."
+
+    return [TextContent(type="text", text=content)]
+
+
+def main() -> None:
+    """Main entry point for the MCP server."""
+    mcp.run()
+
+
+if __name__ == "__main__":
+    main()

{kreuzberg-3.6.1.dist-info → kreuzberg-3.7.0.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: kreuzberg
-Version: 3.6.1
+Version: 3.7.0
 Summary: A text extraction library supporting PDFs, images, office documents and more
 Project-URL: homepage, https://github.com/Goldziher/kreuzberg
 Author-email: Na'aman Hirschfeld <nhirschfed@gmail.com>
@@ -25,7 +25,8 @@ Requires-Python: >=3.10
 Requires-Dist: anyio>=4.9.0
 Requires-Dist: charset-normalizer>=3.4.2
 Requires-Dist: exceptiongroup>=1.2.2; python_version < '3.11'
-Requires-Dist: html-to-markdown>=1.4.0
+Requires-Dist: html-to-markdown[lxml]>=1.6.0
+Requires-Dist: mcp>=1.11.0
 Requires-Dist: msgspec>=0.18.0
 Requires-Dist: playa-pdf>=0.6.1
 Requires-Dist: psutil>=7.0.0
@@ -83,14 +84,15 @@ Description-Content-Type: text/markdown
 
 ## Why Kreuzberg?
 
-- **🚀 Fastest Performance**: [Benchmarked](https://github.com/Goldziher/python-text-extraction-libs-benchmarks) as the fastest text extraction library
-- **💾 Memory Efficient**: 14x smaller than alternatives (71MB vs 1GB+)
+- **🚀 Fastest Performance**: [35+ files/second](https://goldziher.github.io/python-text-extraction-libs-benchmarks/) - the fastest text extraction library
+- **💾 Memory Efficient**: 14x smaller than alternatives (71MB vs 1GB+) with lowest memory usage (~530MB)
 - **⚡ Dual APIs**: Only library with both sync and async support
 - **🔧 Zero Configuration**: Works out of the box with sane defaults
 - **🏠 Local Processing**: No cloud dependencies or external API calls
 - **📦 Rich Format Support**: PDFs, images, Office docs, HTML, and more
 - **🔍 Multiple OCR Engines**: Tesseract, EasyOCR, and PaddleOCR support
-- **🐳 Production Ready**: CLI, REST API, and Docker images included
+- **🤖 AI Integration**: Native MCP server for Claude and other AI tools
+- **🐳 Production Ready**: CLI, REST API, MCP server, and Docker images included
 
 ## Quick Start
 
@@ -136,17 +138,66 @@ asyncio.run(main())
 
 ## Deployment Options
 
+### 🤖 MCP Server (AI Integration)
+
+**Connect directly to Claude Desktop, Cursor, and other AI tools with the Model Context Protocol:**
+
+```bash
+# Install and run MCP server with all features (recommended)
+pip install "kreuzberg[all]"
+kreuzberg-mcp
+
+# Or with uvx (recommended for Claude Desktop)
+uvx --with "kreuzberg[all]" kreuzberg-mcp
+
+# Basic installation (core features only)
+pip install kreuzberg
+kreuzberg-mcp
+```
+
+**Configure in Claude Desktop (`claude_desktop_config.json`):**
+
+```json
+{
+  "mcpServers": {
+    "kreuzberg": {
+      "command": "uvx",
+      "args": ["--with", "kreuzberg[all]", "kreuzberg-mcp"]
+    }
+  }
+}
+```
+
+**Basic configuration (core features only):**
+
+```json
+{
+  "mcpServers": {
+    "kreuzberg": {
+      "command": "uvx",
+      "args": ["kreuzberg-mcp"]
+    }
+  }
+}
+```
+
+**Available MCP capabilities:**
+
+- **Tools**: `extract_document`, `extract_bytes`, `extract_simple`
+- **Resources**: Configuration, supported formats, OCR backends
+- **Prompts**: Extract-and-summarize, structured analysis workflows
+
 ### 🐳 Docker (Recommended)
 
 ```bash
 # Run API server
-docker run -p 8000:8000 goldziher/kreuzberg:3.4.0
+docker run -p 8000:8000 goldziher/kreuzberg:latest
 
 # Extract files
 curl -X POST http://localhost:8000/extract -F "data=@document.pdf"
 ```
 
-Available variants: `3.4.0`, `3.4.0-easyocr`, `3.4.0-paddle`, `3.4.0-gmft`, `3.4.0-all`
+Available variants: `latest`, `3.6.1`, `3.6.1-easyocr`, `3.6.1-paddle`, `3.6.1-gmft`, `3.6.1-all`
 
 ### 🌐 REST API
 
@@ -191,15 +242,20 @@ kreuzberg extract *.pdf --output-dir ./extracted/
 
 ## Performance
 
-**Fastest extraction speeds** with minimal resource usage:
+**[Comprehensive benchmarks](https://goldziher.github.io/python-text-extraction-libs-benchmarks/)** across 94 real-world documents (~210MB) • [View source](https://github.com/Goldziher/python-text-extraction-libs-benchmarks):
+
+| Library       | Speed           | Memory    | Install Size | Dependencies | Success Rate |
+| ------------- | --------------- | --------- | ------------ | ------------ | ------------ |
+| **Kreuzberg** | **35+ files/s** | **530MB** | **71MB**     | **20**       | High\*       |
+| Unstructured  | Moderate        | ~1GB      | 146MB        | 54           | 88%+         |
+| MarkItDown    | Good†           | ~1.5GB    | 251MB        | 25           | 80%†         |
+| Docling       | 60+ min/file‡   | ~5GB      | 1,032MB      | 88           | Low‡         |
 
-| Library       | Speed          | Memory        | Size        | Success Rate |
-| ------------- | -------------- | ------------- | ----------- | ------------ |
-| **Kreuzberg** | ⚡ **Fastest** | 💾 **Lowest** | 📦 **71MB** | ✅ **100%**  |
-| Unstructured  | 2-3x slower    | 2x higher     | 146MB       | 95%          |
-| MarkItDown    | 3-4x slower    | 3x higher     | 251MB       | 90%          |
-| Docling       | 4-5x slower    | 10x higher    | 1,032MB     | 85%          |
+\*_Can achieve 75% reliability with 15% performance trade-off when configured_
+†_Good on simple documents, struggles with large/complex files (>10MB)_
+‡_Frequently fails/times out on medium files (>1MB)_
 
+> **Benchmark details**: Tested across PDFs, Word docs, HTML, images, spreadsheets in 6 languages (English, Hebrew, German, Chinese, Japanese, Korean)
 > **Rule of thumb**: Use async API for complex documents and batch processing (up to 4.5x faster)
 
 ## Documentation
@@ -216,6 +272,7 @@ kreuzberg extract *.pdf --output-dir ./extracted/
 
 ## Advanced Features
 
+- **🤖 MCP Server**: Native integration with Claude Desktop and AI tools
 - **📊 Table Extraction**: Extract tables from PDFs with GMFT
 - **🧩 Content Chunking**: Split documents for RAG applications
 - **🎯 Custom Extractors**: Extend with your own document handlers
@@ -233,7 +290,7 @@ ______________________________________________________________________
 
 <div align="center">
 
-**[Documentation](https://goldziher.github.io/kreuzberg/) • [PyPI](https://pypi.org/project/kreuzberg/) • [Docker Hub](https://hub.docker.com/r/goldziher/kreuzberg) • [Discord](https://discord.gg/pXxagNK2zN)**
+**[Documentation](https://goldziher.github.io/kreuzberg/) • [PyPI](https://pypi.org/project/kreuzberg/) • [Docker Hub](https://hub.docker.com/r/goldziher/kreuzberg) • [Benchmarks](https://github.com/Goldziher/python-text-extraction-libs-benchmarks) • [Discord](https://discord.gg/pXxagNK2zN)**
 
 Made with ❤️ by the [Kreuzberg contributors](https://github.com/Goldziher/kreuzberg/graphs/contributors)
 

{kreuzberg-3.6.1.dist-info → kreuzberg-3.7.0.dist-info}/RECORD

@@ -24,6 +24,8 @@ kreuzberg/_extractors/_pandoc.py,sha256=oQ4DgQSPoX1LXjGAKh_A40JHqiKWb91LeRBYSS_6
 kreuzberg/_extractors/_pdf.py,sha256=R33ggTd0IU6NsEnzgHFTr9ScgcnM8nIIstDq7XMVcvg,14792
 kreuzberg/_extractors/_presentation.py,sha256=ZX-EKQppHwvKtyKk0-IQVF6QAqJi0SfGgCiiyqMQh0w,8701
 kreuzberg/_extractors/_spread_sheet.py,sha256=HOzCeYQc6kaMveAHfi80LrsF0yU7Kn74aKQ7lrMAlo8,6480
+kreuzberg/_mcp/__init__.py,sha256=8PYV-omC8Rln7Cove8C3rHu3d7sR1FuiwSBG1O7vkAE,92
+kreuzberg/_mcp/server.py,sha256=BQHeKI89aKf24BIE4n6m8r1rVA1Zgt6vM8Ki_OHuGnc,6780
 kreuzberg/_multiprocessing/__init__.py,sha256=nwYQpKH7ixHwzkQbTMFCstOCBKktmbNq5dTrwI2Mn94,203
 kreuzberg/_multiprocessing/gmft_isolated.py,sha256=ZfbhiL5bhBEJnibUSls3WV-FECrnU9VvKfq5O2foHcc,11191
 kreuzberg/_multiprocessing/process_manager.py,sha256=_qtB8y9td2coJevlIl4z6F__jau320RdI1lqdyuaeD4,6061
@@ -47,8 +49,8 @@ kreuzberg/_utils/_serialization.py,sha256=AhZvyAu4KsjAqyZDh--Kn2kSWGgCuH7udio8lT
 kreuzberg/_utils/_string.py,sha256=owIVkUtP0__GiJD9RIJzPdvyIigT5sQho3mOXPbsnW0,958
 kreuzberg/_utils/_sync.py,sha256=oT4Y_cDBKtE_BFEoLTae3rSisqlYXzW-jlUG_x-dmLM,4725
 kreuzberg/_utils/_tmp.py,sha256=hVn-VVijIg2FM7EZJ899gc7wZg-TGoJZoeAcxMX-Cxg,1044
-kreuzberg-3.6.1.dist-info/METADATA,sha256=JPTejc7zpahkvhZtUqTVPPVzQ-93aOPnx3l3EQXseok,9160
-kreuzberg-3.6.1.dist-info/WHEEL,sha256=qtCwoSJWgHk21S1Kb4ihdzI2rlJ1ZKaIurTj_ngOhyQ,87
-kreuzberg-3.6.1.dist-info/entry_points.txt,sha256=VdoFaTl3QSvVWOZcIlPpDd47o6kn7EvmXSs8FI0ExLc,48
-kreuzberg-3.6.1.dist-info/licenses/LICENSE,sha256=-8caMvpCK8SgZ5LlRKhGCMtYDEXqTKH9X8pFEhl91_4,1066
-kreuzberg-3.6.1.dist-info/RECORD,,
+kreuzberg-3.7.0.dist-info/METADATA,sha256=0rBXhtDYCdZ2AGpQdXTTZUQUX8T01OsKzwrm2nl14QA,11137
+kreuzberg-3.7.0.dist-info/WHEEL,sha256=qtCwoSJWgHk21S1Kb4ihdzI2rlJ1ZKaIurTj_ngOhyQ,87
+kreuzberg-3.7.0.dist-info/entry_points.txt,sha256=GplGhFryCP7kyAG_k-Mdahznvo2fwi73qLFg5yQfH_A,91
+kreuzberg-3.7.0.dist-info/licenses/LICENSE,sha256=-8caMvpCK8SgZ5LlRKhGCMtYDEXqTKH9X8pFEhl91_4,1066
+kreuzberg-3.7.0.dist-info/RECORD,,

{kreuzberg-3.6.1.dist-info → kreuzberg-3.7.0.dist-info}/entry_points.txt

@@ -1,2 +1,3 @@
 [console_scripts]
 kreuzberg = kreuzberg.cli:cli
+kreuzberg-mcp = kreuzberg._mcp.server:main