Skip to content
CAI
Software that uses CAICheck a score

microsoft/markitdown

59.1

Adequate · 23 October 2026

7.6k

lines of production code

Python

primary language

6

measurements over time

CAI band scale
CAI trend line
CAI lens gauges

What this system is

MarkItDown is a Python library and associated ecosystem designed to convert various file formats into Markdown. It supports a wide range of inputs including Office documents, PDFs, audio, and web pages, utilizing both local parsers and cloud-based services like Azure Content Understanding. The system features a plugin architecture that enables extensibility, such as LLM-based OCR for image text extraction, and provides an MCP server for programmatic access via HTTP or SSE transports.

Features

Add LLM-based OCR service for image text extraction

Introduces a new LLM Vision OCR service that extracts text from images by sending them to OpenAI-compatible vision models (e.g., GPT-4o, Gemini). This allows the tool to process scanned PDFs and embedded images that lack selectable text, returning the extracted content along with confidence metrics and backend details.

markitdown-ocr · high confidence

Add sample RTF converter plugin

Introduces a new sample plugin that enables conversion of Rich Text Format (RTF) files to Markdown. The plugin registers an RtfConverter that accepts .rtf files and RTF MIME types, reads the file stream using the specified or system default encoding, and converts the content to Markdown using the striprtf library.

packages/markitdown-sample-plugin/src · high confidence

Introduce LLM-based OCR plugin for embedded images in Office and PDF files

The new markitdown-ocr plugin (v0.1.1) adds an LLM Vision-based OCR layer that extracts text from images embedded in PDF, DOCX, PPTX, and XLSX files. It registers priority-based converters that intercept image rendering in the core conversion pipelines, sending each image to an LLM (via llm\_client/llm\_model) to produce inline text wrapped in \[Image OCR\]…\[End OCR\] markers, with per-document caching to avoid redundant calls. The plugin requires markitdown\>=0.1.8b3 for the \_image\_to\_html hook and gracefully falls back to native images when OCR fails or returns no text.

packages/markitdown-ocr/src · high confidence

Introduce modular converter package with new file-type support

The converter logic is now organized under the \packages/markitdown/src/markitdown/converters\ package, exposing a centralized \\_\init\\_.py\ that registers all available converters. This update adds dedicated converters for Bing search result pages, CSV files (with improved pipe and newline escaping), EPUBs (with percent-encoded manifest resolution), and audio files (extracting metadata and transcribing speech). It also introduces specialized converters for Azure Document Intelligence and Azure Content Understanding, allowing users to leverage cloud-based extraction for documents, images, and video.

packages/markitdown/src/markitdown/converters · high confidence

Behavioural changes

Introduce StreamInfo dataclass and structured test vectors

The markitdown library now includes a new \StreamInfo\ dataclass to explicitly track file stream metadata such as MIME type, extension, charset, and source path (local or URL). Additionally, a new \FileTestVector\ structure has been added to standardize test definitions, allowing tests to specify expected inclusion/exclusion strings for various file formats (e.g., DOCX, PPTX, PDF) rather than relying on ad-hoc assertions.

markitdown · high confidence

MarkItDown 0.1.8 release with improved URI handling and plugin support

This release updates the package version to 0.1.8 and introduces several behavioral improvements. The CLI now supports the --use-cu flag to route files to Azure Content Understanding, configurable via the MARKITDOWN\_CU\_ENDPOINT environment variable. URI handling has been hardened to reject unsupported Windows UNC and device paths, and data URI parsing now normalizes parameter cases. The library also adds support for loading third-party converters via the plugin entry-point system (enabled with --use-plugins) and improves robustness by increasing the HTTP response read buffer to 64 KB and fixing UTF-8 character splitting during charset detection.

packages/markitdown/src/markitdown · high confidence

MarkItDown MCP server now supports Streamable HTTP transport and plugin configuration

The MarkItDown MCP server has been updated to support the Streamable HTTP transport protocol (preferred over the deprecated SSE transport) and allows users to enable plugins via the MARKITDOWN\_ENABLE\_PLUGINS environment variable. The server now exposes both SSE and Streamable HTTP endpoints, with Streamable HTTP being the default when running in HTTP mode (--http or --sse flags). A security warning is displayed when binding to non-localhost interfaces, and the server version is set to 0.0.1a7.

packages/markitdown-mcp/src · high confidence

Fixes

Improved DOCX and XLSX conversion robustness and feature support

This update introduces several fixes and new capabilities for Office document conversion. It adds support for rendering math equations in .docx files by converting OMML to LaTeX, and enables extraction of images from .xlsx spreadsheets. Reliability is improved by handling DOCX files with inconsistent ZIP filename casing, guarding against missing oMath elements, ignoring malformed styles, and fixing invalid LaTeX macros for symbols like mu, nu, tau, and arrows. The conversion also now correctly preserves strikethrough semantics for double-strike elements and uses defusedxml for secure XML parsing.

_packages/markitdown/src/markitdown/converter\utils · high confidence

Test coverage

Added tests for MCP server transport and error handling; Added tests for the MarkItDown sample plugin's RTF conversion; Added unit tests for markitdown-ocr converters; Consolidated test suite with real fixtures and CLI coverage; Expanded test fixtures for markitdown conversion accuracy.

Dependencies

MarkItDown 0.1.1 release with updated dependencies and new packages

This release updates the core MarkItDown library to version 0.1.1 and introduces three new packages: markitdown-mcp (an MCP server for MarkItDown), markitdown-ocr (an OCR plugin for extracting text from images in PDF, DOCX, PPTX, and XLSX files), and markitdown-sample-plugin (a sample plugin demonstrating the plugin architecture). The core library now depends on magika\~=0.6.1 for file type detection and defusedxml for secure XML parsing. The optional dependencies have been updated to include mammoth\~=1.11.0 for DOCX conversion, pdfminer.six\>=20251230 and pdfplumber\>=0.11.9 for PDF processing, and youtube-transcript-api\~=1.2.3 for YouTube transcription support. The markitdown-ocr package adds support for LLM-based image text extraction via the openai SDK (optional dependency).

(dependencies) · high confidence

Written by watchdog.canine.dev from the codebase's own history, inside the signed delivery this page is composed from.

How this codebase got here

Score

  • CAI 63 → 59 (-4.3)

Lenses

  • Code Health 73 → 72 (-0.2)
  • Architecture 100 → 100 (+0.0)
  • Maturity 67 → 67 (-0.3)
  • Readiness 48 → 50 (+1.3)
  • Security 94 → 57 (-36.6)
  • Performance 100 → 100 (+0.0)

Resolved (3)

  • Documentation: no licence statement (README.md)
  • Off-boarding risk: anonymized user #1
  • PR-triggered workflow without a permissions block

New (10)

  • Duplicated block (12 lines × 2) (packages/markitdown/src/markitdown/_markitdown.py)
  • MarkItDown.enable_builtins (cyclomatic 17) (packages/markitdown/src/markitdown/_markitdown.py)
  • Off-boarding risk: anonymized user #1
  • Orphaned file (packages/markitdown-ocr/src/markitdown_ocr/_ocr_service.py)
  • Orphaned file (packages/markitdown-ocr/src/markitdown_ocr/_pdf_converter_with_ocr.py)
  • Orphaned file (packages/markitdown-ocr/src/markitdown_ocr/_plugin.py)
  • Orphaned knowledge (packages/markitdown/src/markitdown/converters/_pdf_converter.py)
  • PR-triggered workflow without a permissions block (.github/workflows/pre-commit.yml)
  • _doc_intel_converter._field_value (cognitive 19) (packages/markitdown/src/markitdown/converters/_doc_intel_converter.py)
  • _doc_intel_converter._yaml_dump (cognitive 36) (packages/markitdown/src/markitdown/converters/_doc_intel_converter.py)

Changes since last survey

  • 6 commits — 6 feature/other, 0 fixes

By area

  • packages/markitdown — 5 commits
  • packages/markitdown-mcp — 1 commit

Notable commits

  • change: Bump Mammoth dependency to >=1.12.2,<2 (#2729)
  • change: Bump the magika dependency (#2722)
  • change: Bump versions, prior to point-release. (#2733)
  • change: Require certifi>=2026.10.19 for updated CA trust (#2730)
  • change: Update Black, Pillow, and Windows ARM Cryptography baselines (#2731)
  • change: feat(doc-intel): expose models, features, query fields, and structured output (#1913)

Architecture

  • Unchanged — 0 containers · 2 contexts · 0 edges

Written by watchdog.canine.dev from the codebase's own history, inside the signed delivery this page is composed from.

Survey your own repository

microsoft/markitdown was measured the same way every project in this corpus was: the same rubric, at a pinned commit, with the result published in full. Point a surveyor at a repository you know and see whether you agree with it.

About this page

  • The score is its most recent published measurement, taken on 23 October 2026 at a pinned commit. It is not a live figure and does not change until the project is measured again.
  • Measured at commit 7482c092d9c0074a89803c9fdb63a969e7fe17fc — the exact code this score is about.
  • Scored under rubric-2026.10.5 — the same rubric and the same method as every other entry in this index.
  • Measured by watchdog.canine.dev using codehealth-analyzer measure-2026.10.6+preprod-d36780c2d857.