markdownify-mcp

Converts HTML documents to clean Markdown by parsing DOM structure and preserving semantic meaning through intelligent tag mapping. Uses a tree-walking algorithm to traverse HTML nodes and emit corresponding Markdown syntax, handling nested elements, attributes, and special cases like tables, lists, and code blocks. Maintains formatting hierarchy and link references without requiring external HTML-to-Markdown libraries.

Extracts text and structure from PDF documents and converts to Markdown, preserving document hierarchy through detection of headings, sections, and page breaks. Integrates with PDF parsing libraries to extract text layers and metadata, then applies heuristic-based layout analysis to infer Markdown structure (headings, lists, code blocks) from visual positioning and font sizes.

Allows customization of Markdown output format through configuration options (heading style, list markers, link format, code fence style, etc.). Accepts format preferences and applies them consistently across all conversions. Supports multiple Markdown flavors (CommonMark, GitHub Flavored Markdown, Pandoc) with dialect-specific syntax.

Converts images to Markdown by performing OCR on text content and generating natural language descriptions of visual elements. Integrates with OCR engines (Tesseract or cloud APIs) to extract text, then uses vision models or heuristics to describe images, tables, and diagrams, embedding results as Markdown with alt text and code blocks for extracted tables.

Fetches web content from URLs and converts to Markdown in a single operation. Handles HTTP requests with proper headers and redirects, parses HTML responses, and applies HTML-to-Markdown conversion. Includes optional content cleaning (removing navigation, ads, boilerplate) using heuristics or DOM analysis to extract main content before conversion.

Converts structured data (JSON arrays, CSV, database records) into properly formatted Markdown tables. Accepts tabular input, infers column headers and types, and generates Markdown table syntax with proper alignment and escaping. Handles edge cases like null values, long content, and special characters.

Extracts code blocks from documents (HTML, Markdown, plain text) and preserves or infers language syntax highlighting information. Detects code blocks by visual cues (indentation, fencing, monospace fonts) or explicit markers, identifies programming language from context or file extension, and embeds language hints in Markdown code fence syntax.

Extracts metadata (title, author, date, description, tags) from documents and generates Markdown front-matter (YAML or TOML) for use in static site generators or knowledge management systems. Parses HTML meta tags, PDF document properties, and content heuristics to infer metadata, then formats as structured front-matter.

Implements Model Context Protocol server that registers conversion tools as callable functions with JSON schema definitions. Exposes tools to MCP clients (Claude, other LLMs) with input/output schemas, parameter validation, and error handling. Handles tool invocation requests from clients and returns results in MCP-compatible format.

Processes multiple documents in batch mode with progress tracking and error recovery. Accepts a list of documents or URLs, processes each sequentially or in parallel (configurable), tracks progress with callbacks, and handles failures gracefully without stopping the batch. Returns results with per-document status and error details.

Allows composition of multiple conversion steps into custom pipelines (e.g., PDF → HTML → Markdown → table extraction). Provides a pipeline builder API that chains conversion functions, passes output of one step as input to the next, and handles type mismatches or incompatibilities. Supports conditional branching and error recovery within pipelines.