Everything becomes Markdown, and nothing leaves your machine

What it does anymd is a converter with a single output format: Markdown. On the input side it accepts PDF, Word, PowerPoint, Excel, EPUB, HTML and web pages, images (via OCR), and audio/video (metadata, subtitles, transcripts). It comes in two shapes — an MCP server your agent can call as a tool, and a CLI for scripts and humans. It runs on your machine and needs no API key, so there’s no hosted service standing between your files and your agent.

The interesting bit Plenty of tools turn documents into Markdown; this one treats media as a first-class citizen too, OCRing images and pulling subtitles, transcripts, and metadata out of audio and video. It also openly positions itself as a MarkItDown alternative, with Rust where Microsoft’s version has Python — same idea, different engine room.

Key highlights

• Wide ingestion: PDF, DOCX, PPTX, XLSX, EPUB, HTML and live web pages.
• Images go through OCR; audio and video yield metadata, subtitles, and transcripts.
• Two interfaces: an MCP server for agents, a CLI for everything else.
• Local execution, no API key — nothing to sign up for.
• Self-described MarkItDown alternative, written in Rust.

Caveats

• The repo’s README currently documents a different project — the TypeScript npm package @sylphx/pdf-reader-mcp, a PDF-only extraction server — so anymd’s actual options, output quality, and per-format behavior aren’t described in the docs provided.
• The benchmarks in that README (5–10x parallel speedups, ops/sec tables) belong to the predecessor tool, not to anymd’s Rust implementation; don’t carry them over.
• How OCR and transcription are actually implemented — local models, bundled engines, or something else — isn’t stated in any of the sources.

Verdict Anyone wiring an agent or RAG pipeline who’d rather run one local converter than a zoo of per-format parsers should take a look. Anyone who needs well-documented, verifiable behavior today should wait until the README catches up with the repo.