Comparison · Same 38 public documents · Open benchmark

anymd vs Kreuzberg

Both tools are built on Rust, both are MIT licensed, and both turn documents into Markdown. Kreuzberg is a document-intelligence library with bindings for Python, TypeScript, Go, Java, C#, Ruby and more, plus a CLI, a REST API and an MCP server. anymd is one native binary built for AI agents: a CLI and an MCP server that cites pages and reads long files a slice at a time. On the 38 public documents of AgentDocBench, anymd scores 96.4 to Kreuzberg's 81.7, converts all 38 in 11.5 s instead of 16.0 s on the same 4-CPU machine, and returns 19% fewer output tokens. Tables make the biggest difference: table cells F1 is 92.2 against 38.4.

Install anymd free    See the full benchmark

npx -y @sylphx/anymd setup --client cursor,claude-code,codex

At a glance

anymd Kreuzberg
What it is One native Rust binary: CLI and MCP server Rust core with bindings for Python, TypeScript, Rust, Go, Java, C#, Ruby and other languages; CLI, REST API and MCP server
Install npx -y @sylphx/anymd setup, Homebrew, Scoop, Cargo or Docker pip install kreuzberg (Python 3.10 or newer), brew install kreuzberg-dev/tap/kreuzberg, cargo install kreuzberg-cli or Docker
MCP tools Built in: read, outline, search, inspect kreuzberg mcp starts an MCP server
AgentDocBench overall 96.4 81.7
Table cells F1 92.2 38.4
Time, all 38 documents, 4 CPUs 11.5 s 16.0 s
Output tokens, all 38 documents 368,659 455,689
Output Markdown with <!-- page N --> anchors; JSON with geometry through inspect Plain text, Markdown, Djot or HTML, with JSON or TOON output on request
Inputs only Kreuzberg reads — Email (EML, MSG), Korean HWP, archives, source code and more of its 91+ formats
Beyond conversion Page-cited reads, outlines and folder search for agents Chunking, embeddings, language detection and a plugin system
Release line anymd 8, in active development Kreuzberg v4 is a long-term-support line, with critical fixes until the end of 2026 on a best-effort basis; new features go to its successor, Xberg
Licence MIT MIT
Price Free, MIT. Optional anymd Pro, one payment Free

Measured on the same 38 documents

AgentDocBench checks what an agent needs from a converter: the words verbatim, in reading order, with tables intact, fast, and in as few tokens as possible. Scores are 0 to 100; higher is better.

AgentDocBench anymd 8.2.0 Kreuzberg 4.10.4
Overall 96.4 81.7
Documents converted 38/38 38/38
Sentences kept verbatim 97.7 90.1
Text F1 97.2 96.4
Reading order 98.8 96.8
Table cells F1 92.2 38.4
Time, all converted documents 11.5 s 16.0 s
Output tokens, all converted documents 368,659 455,689

Document by document, anymd scores higher on 25, Kreuzberg on 4, and 9 are tied.

By category

Category Documents anymd Kreuzberg
Papers (math) 3 96.7 82.5
Two-column papers 3 98.1 70.5
Reports 2 96.9 83.8
Financial tables 4 95.2 70.6
Forms 3 98.0 73.1
Scanned (OCR) 3 77.9 62.4
CJK 3 99.1 56.9
Slides 4 96.0 82.0
Spreadsheets 4 100.0 100.0
Word 3 99.9 99.8
EPUB 3 98.9 99.1
HTML 3 99.6 100.0

Where Kreuzberg scores higher: EPUB (99.1 vs 98.9); HTML (100.0 vs 99.6); paper-rankone (98.0 vs 94.8); scan-nasa-notes (66.7 vs 62.1); epub-markitdown (97.2 vs 96.8); html-bls-ooh-software-developers (100.0 vs 98.7).

Where Kreuzberg shines: it reads HTML and EPUB slightly better than anymd here, ties it on spreadsheets at 100.0, and is within a point on Word and on text F1 (96.4 vs 97.2). The gap opens on structure: tables, two-column papers, CJK, forms and financial tables.

How it was measured

Both tools ran on GitHub-hosted Linux x86_64 runners with 4 CPUs: Kreuzberg 4.10.4, the latest v4 release, through its Python API (extract_file_sync) with Markdown output and the cache off on 2026-09-25, and anymd 8.2.0 on 2026-10-01. Kreuzberg's default configuration leaves OCR off, so the run turned on its Tesseract OCR in English to read scanned pages, as the other tools do by default; pages with a text layer were not OCR'd. Time is the wall time of a fresh process per file, start-up included: the median of three runs after a warm-up for both tools. Kreuzberg's other OCR backends and its successor Xberg were not measured. Tokens are counted with o200k_base.

The benchmark lives in the anymd repository. The scoring does not know which tool produced an output, and the corpus, ground truth, scoring and raw results are public in bench/. Anyone can rerun it or submit a tool.

Why agent builders pick anymd

Tables an LLM can read

anymd builds tables from the PDF's drawn lines and aligned columns, so cells stay in their rows: table cells F1 is 92.2 against Kreuzberg's 38.4. Financial tables score 95.2 against 70.6, and forms 98.0 against 73.1.

Columns and CJK stay in order

anymd reads glyph positions and finds the gutters between columns, so two-column papers score 98.1 against 70.5 and Chinese and Japanese documents 99.1 against 56.9. More sentences survive verbatim: 97.7 against 90.1.

Faster on the whole suite, in fewer tokens

On the same 4-CPU runner, anymd converted the 38 documents in 11.5 s and Kreuzberg in 16.0 s, and anymd returned 368,659 output tokens to Kreuzberg's 455,689. Most of both totals is OCR on the three scanned PDFs.

An MCP server built for long files

npx -y @sylphx/anymd setup adds anymd to Claude Code, Codex, Cursor, VS Code, Claude Desktop, Windsurf and Gemini CLI at once. The MCP tools add what an agent needs for long files: <!-- page N --> anchors to cite, outline to jump to a section, read with a token budget and a cursor, and search across a folder. Files stay on your machine, and --allow-dir keeps the MCP server inside one folder.

When Kreuzberg is the better fit

  • You want an in-process library in Python, TypeScript, Go, Java, C#, Ruby or another of its bindings. anymd's Python package is prepared but not yet on PyPI.
  • You read email, Korean HWP, archives or source code, which anymd does not read.
  • You want chunking, embeddings and language detection in the same library as extraction.
  • Your corpus is mostly HTML or EPUB, where Kreuzberg scored slightly higher on this benchmark.

Switch in one line

kreuzberg extract report.pdf --content-format markdown   # Kreuzberg
anymd report.pdf -o report.md                            # anymd, after npm install -g @sylphx/anymd

For MCP, replace the kreuzberg mcp entry in your client with npx -y @sylphx/anymd, or run npx -y @sylphx/anymd setup to add it to every client. With Docker:

docker run --rm -v "$PWD:/data" ghcr.io/sylphxai/anymd report.pdf > report.md

More install options are in Getting started.

Check what your agent cites with anymd Pro

Converting a file is free. When an agent quotes a PDF or a video, anymd Pro takes you back to the source:

  • Cite-check compares up to 100 quotes per request with the page and area each one cites, and reports whether it matches, doesn't match, or needs more evidence. It uses the text and its position on the page, not an AI opinion.
  • Video timelines and frames show scenes, chapters and subtitle or transcript timings, and the frame behind a cited moment with its actual decoded timestamp and a SHA-256 hash of the image.

One-time US$29 purchase. No subscription. Sign in, pay securely, then run anymd login with the same account to activate Pro on your machine. After activation, Pro works offline.

See Anymd Pro

FAQ

Is anymd a drop-in replacement for Kreuzberg? From the command line or an MCP client, mostly: anymd report.pdf -o report.md replaces kreuzberg extract report.pdf --content-format markdown, and npx -y @sylphx/anymd setup adds anymd to your MCP clients. Code that calls Kreuzberg as a library cannot switch today, because anymd's Python package is not on PyPI yet.

Which one is more accurate? On AgentDocBench, anymd scores 96.4 overall and Kreuzberg 81.7. anymd scores higher on 25 documents and Kreuzberg on 4, and 9 are tied. The largest gap is tables, with table cells F1 of 92.2 against 38.4. Kreuzberg scores slightly higher on HTML and EPUB.

Is anymd faster than Kreuzberg? On AgentDocBench, anymd converted all 38 documents in 11.5 seconds and Kreuzberg in 16.0 seconds on the same 4-CPU runner, each timed as a fresh process per file with start-up included.

Does this comparison cover Xberg? No. It measures Kreuzberg 4.10.4, the latest release of the v4 line. Kreuzberg's maintainers now add new features to its successor, Xberg, which was not measured. Anyone can add it to the benchmark with one adapter file.

Is the benchmark biased toward anymd? It lives in the anymd repository, so you can check it: the scoring does not know which tool produced an output, the corpus and ground truth are public, and the categories where anymd loses are published. Anyone can rerun it or add a tool.

Is anymd free? Yes. anymd is MIT licensed, including the CLI, the MCP server, OCR and transcripts. anymd Pro is an optional one-time purchase, US$29 or the local price in 16 currencies, that adds PDF cite-check and video timelines and frames.

Kreuzberg is an open-source project of its own maintainers; anymd is not affiliated with or endorsed by the Kreuzberg or Xberg projects. Benchmark figures come from the AgentDocBench results committed in the anymd repository. Kreuzberg details come from its documentation.