anymd

Any file → clean Markdown for AI agents.

PDF, Word, PowerPoint, Excel, EPUB, HTML, images, audio/video. A fast Rust engine running on your machine, available as an MCP server and a CLI. No API key.

Install · Benchmarks · Tools · CLI · Formats · Docs

Formerly pdf-reader-mcp / Citra. @sylphx/pdf-reader-mcp and @sylphx/citra still install and run anymd.

Why anymd

• Fast. Native Rust converts in parallel, page by page. A 15-page paper converts in about 0.1 s. That is 20× faster than MarkItDown and 500× faster than docling, with every table intact.
• Accurate. A layout engine rebuilds words from glyph gaps, puts two-column papers in reading order, and recovers tables, including borderless ones. The text stays exactly as printed, with no glued words and no scrambled columns.
• Lean on tokens. Pages come back as Markdown with <!-- page 3 --> citation anchors, a small front-matter header, and compact tables. A token budget and a cursor keep large documents within your agent's context.
• Every format, one call. One tool reads every format listed below. It also accepts web URLs and whole directories, and search looks across all of them.
• Local and private. Nothing is uploaded. OCR and transcripts use local tools you already have (tesseract, ffmpeg, whisper.cpp), and only when they are installed.

Install

Every MCP client runs the same command, npx -y @sylphx/anymd. Node 18+ is the only requirement; npm installs the native binary for your platform.

claude mcp add anymd -- npx -y @sylphx/anymd

codex mcp add anymd -- npx -y @sylphx/anymd

or in ~/.codex/config.toml:

[mcp_servers.anymd]
command = "npx"
args = ["-y", "@sylphx/anymd"]

or in .cursor/mcp.json:

{ "mcpServers": { "anymd": { "command": "npx", "args": ["-y", "@sylphx/anymd"] } } }

code --add-mcp '{"name":"anymd","command":"npx","args":["-y","@sylphx/anymd"]}'

or in .vscode/mcp.json:

{ "servers": { "anymd": { "type": "stdio", "command": "npx", "args": ["-y", "@sylphx/anymd"] } } }

Add to claude_desktop_config.json (Settings → Developer → Edit Config):

{ "mcpServers": { "anymd": { "command": "npx", "args": ["-y", "@sylphx/anymd"] } } }

Any client that speaks MCP over stdio: command npx, args ["-y", "@sylphx/anymd"]. To keep the server inside one folder, add --allow-dir=/path/to/docs.

npm install -g @sylphx/anymd     # or run it once with: npx -y @sylphx/anymd <file>

Benchmarks

Twelve real documents (papers, a two-column paper, statistical tables, CJK, a form, a borderless-table invoice, plus DOCX, PPTX, XLSX, EPUB, and a Wikipedia page), run on a GitHub-hosted runner with 4 CPUs:

¹ pdftotext reads PDFs only and outputs plain text without tables.

On the 15-page Attention Is All You Need paper, anymd takes 0.13 s, docling 76 s, and MarkItDown 3.0 s, and MarkItDown glues the words together ("dominantsequencetransductionmodels"). On the Wikipedia article, anymd's main-content extraction uses 21.7k tokens; docling uses 37.6k, kreuzberg 51.7k, and MarkItDown 54.6k.

Each tool runs as a fresh process, and every number is the median of 3 runs (docling runs once, after its models are warmed up). Tokens are counted with o200k_base. Sentences intact counts reference sentences that come out verbatim; glued words or split columns fail the check. Table rows counts ground-truth rows that come out as one Markdown table row with the cells in order. The method, corpus, ground truth, raw results, and scripts are in bench/, and the Benchmark workflow re-runs everything on GitHub-hosted runners.

The official @modelcontextprotocol/server-pdf is left out of the table because it has no headless text path. It renders PDFs in an interactive viewer, and its read_pdf_bytes tool returns base64-encoded bytes.

MCP tools

anymd exposes three tools.

A read answer looks like this:

---
source: papers/attention.pdf
title: Attention Is All You Need
pages: 15
showing: pages 1-9
---

<!-- page 1 -->

# Attention Is All You Need
…

<!-- page 8 -->

|Model|BLEU EN-DE|BLEU EN-FR|
|-|-|-|
|Transformer (big)|28.4|41.8|
…

<!-- Stopped at the 20000-token budget. Continue with cursor: "10", or pick pages, or raise max_tokens. -->

search answers with one line per hit:

5 matches for "masked language model" (2 files, 31 sections searched)

### papers/bert.pdf (5)
- p.1: …by using a “**masked language model**” (MLM) pre-training objective, inspired by the Cloze task…
- p.2: …In addition to the **masked language model**, we also use a “next sentence prediction” task…

If nothing matches exactly, search falls back to BM25-ranked passages, so a question like "how does bidirectional pretraining work" still finds the right page.

The pdf-reader-mcp tool names (read_pdf, search_pdf, pdf_evidence, pdf_compare) still work for this major version. They no longer appear in tools/list.

CLI

The same binary is a command-line converter, like MarkItDown but much faster:

anymd report.pdf > report.md                 # a file
anymd deck.pptx notes.docx budget.xlsx        # several files, each with a header
anymd https://example.com/article            # a web page (main content only)
cat scan.png | anymd - --ocr                 # stdin, with OCR
anymd paper.pdf --pages 1-3 --max-tokens 4000
anymd search "indemnification" contracts/ --glob '*.pdf'
anymd doctor                                 # lists the optional tools anymd found

Run with no arguments from an MCP client (piped stdin), or as anymd mcp, and it serves MCP over stdio.

Formats

How it works

For PDFs, anymd reads glyph positions rather than text runs. Glyphs are grouped into lines by baseline, which tolerates super- and subscripts. Word spaces come from the gaps between glyphs, measured against the font size and adjusted for letter tracking. A column-aware XY cut finds gutters between running text. Rows whose cells line up become pipe