# Markitai

> Opinionated Markdown converter with native LLM enhancement support. Converts documents, images, and URLs to clean Markdown, with optional LLM cleaning, frontmatter metadata, vision analysis, OCR for scans, and a local web workspace.

Command-line tool (`markitai`, alias `mkai`) and Python package. Source and issue tracker: https://github.com/Ynewtime/markitai. Docs are bilingual: English at the site root, 简体中文 under `/zh/`.

License: markitai's own source code is MIT. The default install is not uniformly MIT — the PDF engine is PyMuPDF (`pymupdf`, `pymupdf-layout`, `pymupdf4llm`) from Artifex Software, dual-licensed AGPL-3.0 or commercial. Local use is unaffected; redistributing the combined work or offering it to others over a network triggers AGPL-3.0 obligations (or requires a commercial licence from Artifex). See NOTICE for full third-party attribution: https://github.com/Ynewtime/markitai/blob/main/NOTICE

## Automation

- `markitai input -o out/ --json` returns final per-item status, output paths, usage and costs; check the exit code and `ok`/`error`. `--dry-run` and `--llm-batch-collect` exclude JSON output.
- `--no-remote-fetch` disables remote URL extraction; it does not disable separately enabled LLM enhancement or a Cloudflare file backend.
- MCP conversions accept `profile`; batches accept `concurrency` (default 10) and isolate outputs by batch/item. Omitted `llm` inherits server configuration; `llm: false` disables it.

## Docs

- [Getting Started](https://markitai.dev/guide/getting-started): Install (guided script, uv, pipx), prerequisites, first conversion, output structure, supported formats
- [Why Markitai](https://markitai.dev/guide/comparison): How markitai compares with markitdown, docling and anydoc, and which job each tool optimizes for
- [CLI Reference](https://markitai.dev/guide/cli): Every command and flag
- [Output Profiles](https://markitai.dev/guide/output-profiles): rag/obsidian/okf output shaping, frozen images.json schema, LlamaIndex ingestion recipe
- [Python API](https://markitai.dev/guide/python-api): Programmatic `convert`/`aconvert` with typed results (provisional)
- [MCP Server](https://markitai.dev/guide/mcp): bundled `markitai-mcp` server (`markitai[mcp]` extra) exposing conversion tools to AI agents over stdio
- [Configuration](https://markitai.dev/guide/configuration): Config file and env vars, LLM providers, image/screenshot/OCR/office settings, cache, security
- [Fetch Policy](https://markitai.dev/guide/fetch-policy): URL fetch strategy cascade (static/Playwright/Defuddle/Jina/Cloudflare), domain profiles, remote-fetch consent
- [Web Workspace](https://markitai.dev/guide/serve): `markitai serve` local UI, job history, REST + SSE API
- [Changelog](https://markitai.dev/changelog): Release history

## 中文文档

- [快速开始](https://markitai.dev/zh/guide/getting-started): 安装、前置依赖、首次转换、输出结构、支持格式
- [为什么选 Markitai](https://markitai.dev/zh/guide/comparison): 与 markitdown、docling、anydoc 的对比，以及每个工具各自优化的目标
- [CLI 命令](https://markitai.dev/zh/guide/cli): 完整命令与参数参考
- [输出 Profile](https://markitai.dev/zh/guide/output-profiles): rag/obsidian/okf 输出塑形、冻结的 images.json schema、LlamaIndex 摄取示例
- [Python API](https://markitai.dev/zh/guide/python-api): 程序化调用 `convert`/`aconvert` 与类型化结果（暂定阶段）
- [MCP 服务器](https://markitai.dev/zh/guide/mcp): 随主包发布的 `markitai-mcp` 服务器（`markitai[mcp]` extra），通过 stdio 向 AI Agent 提供转换工具
- [配置说明](https://markitai.dev/zh/guide/configuration): 配置文件与环境变量、LLM 提供商、缓存、安全设置
- [抓取策略](https://markitai.dev/zh/guide/fetch-policy): URL 抓取策略级联、域名配置、远程抓取同意
- [网页工作台](https://markitai.dev/zh/guide/serve): `markitai serve` 本地界面、历史记录、REST + SSE API
- [更新日志](https://markitai.dev/zh/changelog): 发布历史

## Optional

- [GitHub repository](https://github.com/Ynewtime/markitai): Source, issues, NOTICE and license details
- [PyPI package](https://pypi.org/project/markitai/): Installable releases