Skip to content
Cobalt Rentals

MCP Registry

MarkItDown

PyPI
PyPI - Downloads

Important

MarkItDown performs I/O with the privileges of the current process. Like open() or requests.get(), it will access resources that the process itself can access. Sanitize your inputs in untrusted environments, and call the narrowest convert_* function needed for your use case (e.g., convert_stream(), or convert_local()). See the Security Considerations section of the documentation for more information.

MarkItDown is a lightweight Python utility for converting various files to Markdown for use with LLMs and related text analysis pipelines. To this end, it is most comparable to textract, but with a focus on preserving important document structure and content as Markdown (including: headings, lists, tables, links, etc.) While the output is often reasonably presentable and human-friendly, it is meant to be consumed by text analysis tools -- and may not be the best option for high-fidelity document conversions for human consumption.

MarkItDown currently supports the conversion from:

  • PDF
  • PowerPoint
  • Word
  • Excel
  • Images (EXIF metadata and OCR)
  • Audio (EXIF metadata and speech transcription)
  • HTML
  • Text-based formats (CSV, JSON, XML)
  • ZIP files (iterates over contents)
  • YouTube URLs
  • EPubs
  • ... and more!

Why Markdown?

Markdown is extremely close to plain text, with minimal markup or formatting, but still
provides a way to represent important document structure. Mainstream LLMs, such as
OpenAI's GPT-4o, natively "speak" Markdown, and often incorporate Markdown into their
responses unprompted. This suggests that they have been trained on vast amounts of
Markdown-formatted text, and understand it well. As a side benefit, Markdown conventions
are also highly token-efficient.

Prerequisites

MarkItDown requires Python 3.10 through 3.14. It is recommended to use a virtual environment to avoid dependency conflicts.

With the standard Python installation, you can create and activate a virtual environment using the following commands:

python -m venv .venv source .venv/bin/activate

If using uv, you can create a virtual environment with:

uv venv --python=3.12 .venv source .venv/bin/activate

NOTE: Be sure to use 'uv pip install' rather than just 'pip install' to install packages in this virtual environment

If you are using Anaconda, you can create a virtual environment with:

conda create -n markitdown python=3.12 conda activate markitdown

Installation

To install MarkItDown, use pip: pip install 'markitdown[all]'. Alternatively, you can install it from the source:

git clone git@github.com:microsoft/markitdown.git cd markitdown pip install -e 'packages/markitdown[all]'

Usage

Command-Line

markitdown path-to-file.pdf > document.md

Or use -o to specify the output file:

markitdown path-to-file.pdf -o document.md

You can also pipe content:

cat path-to-file.pdf | markitdown

Optional Dependencies

MarkItDown has optional dependencies for activating various file formats. Earlier in this document, we installed all optional dependencies with the [all] option. However, you can also install them individually for more control. For example:

pip install 'markitdown[pdf, docx, pptx]'

will install only the dependencies for PDF, DOCX, and PPTX files.

At the moment, the following optional dependencies are available:

  • [all] Installs all optional dependencies
  • [pptx] Installs dependencies for PowerPoint files
  • [docx] Installs dependencies for Word files
  • [xlsx] Installs dependencies for Excel files
  • [xls] Installs dependencies for older Excel files
  • [pdf] Installs dependencies for PDF files
  • [outlook] Installs dependencies for Outlook messages
  • [az-doc-intel] Installs dependencies for Azure Document Intelligence
  • [az-content-understanding] Installs dependencies for Azure Content Understanding
  • [audio-transcription] Installs dependencies for audio transcription of wav and mp3 files
  • [youtube-transcription] Installs dependencies for fetching YouTube video transcription

Plugins

MarkItDown also supports 3rd-party plugins. Plugins are disabled by default. To list installed plugins:

markitdown --list-plugins

To enable plugins use:

markitdown --use-plugins path-to-file.pdf

To find available plugins, search GitHub for the hashtag #markitdown-plugin. To develop a plugin, see packages/markitdown-sample-plugin.

markitdown-ocr Plugin

The markitdown-ocr plugin adds OCR support to PDF, DOCX, PPTX, and XLSX converters, extracting text from embedded images using LLM Vision — the same llm_client / llm_model pattern that MarkItDown already uses for image descriptions. No new ML libraries or binary dependencies required.

Installation:

pip install markitdown-ocr pip install openai # or any OpenAI-compatible client

Usage:

Pass the same llm_client and llm_model you would use for image descriptions:

from markitdown import MarkItDown from openai import OpenAI

md = MarkItDown( enable_plugins=True, llm_client=OpenAI(), llm_model="gpt-4o", ) result = md.convert("document_with_images.pdf") print(result.markdown)

If no llm_client is provided the plugin still loads, but OCR is silently skipped and the standard built-in converter is used instead.

See packages/markitdown-ocr/README.md for detailed documentation.

Azure Content Understanding

Azure Content Understanding provides higher-quality conversion with structured field extraction (YAML front matter), multi-modal support (documents, images, audio, video), and configurable analyzers.

Install: pip install 'markitdown[az-content-understanding]'

When to use Content Understanding

Content Understanding is ideal when you need capabilities beyond what built-in or Document Intelligence converters provide:

  • Audio and video files — CU is the only option for video, and the higher-quality cloud option for audio. Built-in converters have no video support and only basic audio transcription.

  • Structured field extraction — Prebuilt or custom-built analyzers extract domain-specific fields (invoice amounts, receipt dates, contract clauses) serialized as YAML front matter. Neither built-in nor Doc Intel integration exposes fields.

  • Higher-quality document extraction — Cloud-based layout analysis and OCR for scanned PDFs, complex tables, and multi-page documents.

  • Single API for all modalities — One cu_endpoint handles documents, images, audio, and video with automatic analyzer routing.

  • Built-in converters

    • Document conversion: Offline, format-specific extraction
    • Structured fields: Not available
    • Custom analyzers/models: Not available
    • Audio and video: Basic audio, no video
    • Cost: Local compute only

    Azure Document Intelligence

    • Document conversion: Cloud layout extraction
    • Structured fields: YAML front matter from model or query fields
    • Custom analyzers/models: Supported with docintel_model_id
    • Audio and video: Not supported
    • Cost: Billable Azure API calls

    Azure Content Understanding

    • Document conversion: Cloud multimodal extraction
    • Structured fields: YAML front matter from analyzer fields
    • Custom analyzers/models: Supported with cu_analyzer_id
    • Audio and video: Audio and video analyzers
    • Cost: Billable Azure API calls

    CLI:

    markitdown path-to-file.pdf --use-cu --cu-endpoint "<content_understanding_endpoint>"

    The endpoint can also be set once in the environment, so callers only need --use-cu:

    export MARKITDOWN_CU_ENDPOINT="<content_understanding_endpoint>" markitdown path-to-file.pdf --use-cu

    Python API:

    from markitdown import MarkItDown

    # Zero-config — auto-selects analyzer per file type md = MarkItDown(cu_endpoint="<content_understanding_endpoint>") result = md.convert("report.pdf") # documents → prebuilt-documentSearch result = md.convert("meeting.mp4") # video → prebuilt-videoSearch result = md.convert("call.wav") # audio → prebuilt-audioSearch print(result.markdown)

    With a custom analyzer (for domain-specific field extraction):

    md = MarkItDown( cu_endpoint="<content_understanding_endpoint>", cu_analyzer_id="my-invoice-analyzer", ) result = md.convert("invoice.pdf") print(result.markdown) # Output includes YAML front matter with extracted fields: # --- # contentType: document # fields: # VendorName: CONTOSO LTD. # InvoiceDate: '2019-11-15' # --- # # ...

    When cu_analyzer_id is set, the converter automatically scopes it to compatible file types based on the analyzer's modality. Incompatible types (e.g., audio files with a document analyzer) auto-route to default prebuilt analyzers.

    Cost note: Each convert() call for a CU-routed format is a billable Azure API call. Use cu_file_types to restrict which formats route to CU:

    from markitdown.converters import ContentUnderstandingFileType

    md = MarkItDown( cu_endpoint="<content_understanding_endpoint>", cu_file_types=[ContentUnderstandingFileType.PDF], # only PDFs use CU )

    More information about Azure Content Understanding can be found here.

    Azure Document Intelligence

    To use Microsoft Document Intelligence for conversion:

    export AZURE_API_KEY="<document_intelligence_api_key>" markitdown path-to-file.pdf -o document.md -d -e "<document_intelligence_endpoint>"

    The endpoint can also be set once in the environment, so callers only need -d:

    export MARKITDOWN_DOCINTEL_ENDPOINT="<document_intelligence_endpoint>" markitdown path-to-file.pdf -o document.md -d

    If AZURE_API_KEY is not set, MarkItDown uses DefaultAzureCredential.

    Select a model, configure analysis add-ons, or request fields from the CLI:

    markitdown document.pdf --use-docintel \ --endpoint "<document_intelligence_endpoint>" \ --docintel-model-id prebuilt-layout \ --docintel-features "ocrHighResolution,styleFont" \ --docintel-query-fields "Title,Author"

    No analysis add-ons are enabled by default. Use --docintel-features to opt in,
    or --docintel-features none to explicitly request no add-ons. Analysis add-ons
    and query fields apply only to OCR-supported inputs such as PDFs and images.

    Available feature values are barcodes, formulas, keyValuePairs,
    languages, ocrHighResolution, and styleFont. The formulas,
    ocrHighResolution, and styleFont features incur additional charges, as does
    the queryFields add-on, which is priced differently from the other add-ons.
    See the
    Document Intelligence add-on documentation
    and pricing
    for details.

    More information about how to set up an Azure Document Intelligence Resource can be found here

    Python API

    Basic usage in Python:

    from markitdown import MarkItDown

    md = MarkItDown(enable_plugins=False) # Set to True to enable plugins result = md.convert("test.xlsx") print(result.markdown)

    Document Intelligence conversion in Python using an API key:

    from azure.core.credentials import AzureKeyCredential from markitdown import MarkItDown

    md = MarkItDown( docintel_endpoint="<document_intelligence_endpoint>", docintel_credential=AzureKeyCredential("<document_intelligence_api_key>"), ) result = md.convert("test.pdf") print(result.markdown)

    Omit docintel_credential to use AZURE_API_KEY when it is set, or
    DefaultAzureCredential otherwise.

    Configure the model, analysis add-ons, and structured query fields:

    from azure.ai.documentintelligence.models import DocumentAnalysisFeature from markitdown import MarkItDown

    md = MarkItDown( docintel_endpoint="<document_intelligence_endpoint>", docintel_model_id="prebuilt-layout", docintel_features=[ DocumentAnalysisFeature.OCR_HIGH_RESOLUTION, DocumentAnalysisFeature.STYLE_FONT, ], docintel_query_fields=["Title", "Author"], ) result = md.convert("test.pdf") print(result.markdown)

    Omit docintel_features or set docintel_features=[] to request no analysis
    add-ons. When query fields are provided, MarkItDown automatically enables the
    required queryFields add-on.
    Models that return structured fields prepend them to the Markdown as YAML front
    matter. A single document uses a top-level fields mapping. When the service
    returns multiple documents, the front matter instead contains a documents
    list in service order, with a separate fields mapping for each document.
    Documents without extracted fields retain an empty mapping in that list; no
    front matter is emitted when all documents have empty fields. Selection groups
    are represented as lists of selected values.

    To use Large Language Models for image descriptions (currently only for pptx and image files), provide llm_client and llm_model:

    from markitdown import MarkItDown from openai import OpenAI

    client = OpenAI(max_retries=5) md = MarkItDown(llm_client=client, llm_model="gpt-4o", llm_prompt="optional custom prompt") result = md.convert("example.jpg") print(result.markdown)

    max_retries controls the OpenAI client's automatic retries for retryable errors (the default is 2); 5 allows up to six attempts with backoff. See the OpenAI SDK retry documentation.

    If any attempt succeeds, image conversion continues normally. If the client raises an error after exhausting its retries, or encounters a non-retryable error, MarkItDown tries other applicable converters and raises FileConversionException only if none succeeds.

    Docker

    docker build -t markitdown:latest . docker run --rm -i markitdown:latest < ~/your-file.pdf > output.md