Free MIT Rust MCP server & CLI · No API key

Any file to clean Markdown for AI agents—locally in milliseconds.

Python document converters take minutes to load models or scramble multi-column PDFs and tables. anymd is a native Rust binary and MCP server that parses PDF, Word, PowerPoint, Excel, EPUB, HTML, OCR images, and audio/video directly on your machine—scoring 96.4 across 38 documents in 11.5 seconds on AgentDocBench.

Install via npx -y @sylphx/anymd setup    See AgentDocBench results

Why teams switch to anymd

1. One command wires Claude Code, Codex, Cursor, VS Code & Claude Desktop

Run npx -y @sylphx/anymd setup (or install via cargo install anymd, brew install sylphxai/tap/anymd, Scoop from the SylphxAI/scoop-bucket bucket, or Docker ghcr.io/sylphxai/anymd). Zero Python virtualenvs, zero LibreOffice installs, and zero API keys.

2. Reading-order Markdown and intact tables without burning context tokens

anymd rebuilds words from PDF glyph gaps, orders two-column papers correctly, recovers borderless tables, strips running headers, and adds <!-- page N --> anchors. Use outline to jump straight to a section and read with a token budget so huge files stay inside context.

3. 100% local privacy—plus verifiable PDF quote checks and video frames with Pro

Files stay on your machine (--allow-dir locks MCP scope to one folder). Need deterministic PDF quote verification (cite_check) or SHA-256 verifiable video frames? anymd Pro unlocks both offline in the same binary for a one-time US$29 payment. See Local core and anymd Pro.

Measured on AgentDocBench (38 Docs / 12 Categories)

Converter Overall Score Docs Converted Table Cells F1 Time (All 38 Docs) Output Tokens
anymd (Rust, local MIT) 96.4 38 / 38 92.2 11.5 s 368,659
Docling 93.0 38 / 38 89.9 2,432.4 s 430,148
Kreuzberg 81.7 38 / 38 38.4 16.0 s 455,689
Unstructured 81.2 38 / 38 38.4 346.5 s 414,333
MarkItDown 76.8 38 / 38 57.2 75.6 s 538,388
Marker 71.0 30 / 38 60.9 7,104.5 s 277,779
pdftotext 42.2 23 / 38 0.0 0.90 s 201,448

Coming from MarkItDown? See anymd vs MarkItDown for a side-by-side comparison and a one-line switch. Comparing Python pipelines? See anymd vs Docling, Marker or Unstructured.

FAQ

What file formats does anymd convert with zero external cloud calls? One read call handles PDF, Word (.docx), PowerPoint (.pptx), Excel/OpenDocument (.xlsx, .xls, .ods, CSV/TSV), EPUB, HTML and URLs, images (EXIF + OCR via local doc-VLM or Tesseract), and audio/video metadata, subtitles, and transcripts.

Does anymd require an API key or upload my documents? No. Everything runs locally on your machine under the MIT licence, and no network call is made unless you explicitly pass an http(s) URL.

Why is anymd faster than Docling, Marker, or MarkItDown? anymd is compiled native Rust (anymd-core, anymd-pdf, anymd-formats) that parses XML/ZIP and PDF glyph coordinates in parallel page by page without loading heavy Python ML frameworks for native text documents.

Is the core MCP server free forever, or gated behind Pro? Reading, outlining, searching, OCR, transcripts, the CLI, and all four core MCP tools (read, outline, search, inspect) remain 100% free and MIT licensed. Only offline cite_check and verifiable video frame/timeline inspection require anymd Pro ($29 one-time).

Need to verify PDF quotes or cite exact video frames? See Anymd Pro.