AGENT SKILL

Turn PDFs into Markdown your agent can use

The pdf-to-markdown skill converts PDFs on your computer and preserves headings, lists, tables, and reading order. Use it with Claude, Codex, or another agent.

Structured for agents

Preserves headings, lists, tables, and reading order instead of returning a block of plain text.

Your PDFs stay on your computer

Your PDFs and converted files aren’t uploaded to Nutrient.

Standard is free

Convert PDFs that contain selectable text without an account. Use Vision for scans, handwriting, and complex layouts.

INSTALL

Install the skill or the CLI

Add the skill to a supported agent, or install the CLI for direct use in a terminal or script.

/

Install the agent skill

The Skills CLI installs pdf-to-markdown in Claude Code, Codex, Cursor, Gemini CLI, and other compatible agents.

npx skills add \
pspdfkit-labs/nutrient-skills \
--skill pdf-to-markdown

Install the standalone CLI

Install the CLI from npm on macOS with Apple Silicon, Linux, or Windows. Then run nutrient --version to check the installation. Compare Vision plans.

npm install -g @pspdfkit/pdf-to-markdown

COMMANDS

Convert to Markdown or text. Then search.

All three commands run on your computer. Standard conversion and search are free.


pdf-to-markdown

Create structured Markdown

Preserve headings, lists, tables, and reading order for agents and search systems.


pdf-to-text

Create plain text

Keep columns and spacing when Markdown isn’t the right output format.


query

Search an extracted file

Return the passages that best match a query without converting the PDF again.


BENCHMARKS

Choose speed or accuracy without changing tools

Standard is tied for fastest in the benchmark. Vision improves accuracy for difficult pages and still runs faster than Docling.

Nutrient Other tools
Where Nutrient lands on quality
Vision reading order
0.96 1st of 8 configurations
Vision extraction accuracy
0.93 1st of 8 configurations
Vision table structure
0.94 1st of 8 configurations
ACCURACY ↑ Higher is better · Score 0–1
Extraction accuracy
Nutrient Vision 0.93
Docling 0.89
Nutrient Standard 0.89
pymupdf4llm 0.86
OpenDataLoader 0.83
markitdown 0.59
pypdf 0.58
liteparse 0.57
Reading order
Nutrient Vision 0.96
Nutrient Standard 0.93
Docling 0.91
pymupdf4llm 0.90
OpenDataLoader 0.90
pypdf 0.87
liteparse 0.86
markitdown 0.84
Table structure
Nutrient Vision 0.94
Docling 0.93
Nutrient Standard 0.74
pymupdf4llm 0.73
OpenDataLoader 0.48
markitdown 0.27
liteparse 0.00
pypdf 0.00
Heading levels
Nutrient Vision 0.87
Docling 0.83
Nutrient Standard 0.82
pymupdf4llm 0.78
OpenDataLoader 0.74
markitdown 0.00
liteparse 0.00
pypdf 0.00
SPEED ↓ Lower is better · Seconds per page
Extraction time per page Seconds · Log scale
Nutrient Standard 0.004 s
liteparse 0.004 s
OpenDataLoader 0.015 s
pypdf 0.015 s
markitdown 0.069 s
pymupdf4llm 0.218 s
Nutrient Vision 0.354 s
Docling 0.549 s

134×
Standard faster than Docling
53×
Standard faster than pymupdf4llm
17×
Standard faster than markitdown
4×
Standard faster than pypdf
4×
Standard faster than OpenDataLoader
1×
Standard comparable to liteparse

Standard was measured on 6 July 2026, and Vision was measured on 7 July 2026, both using an Apple M3 Ultra without GPU acceleration. See the GitHub repository for methodology and the latest runs.

The benchmark uses 200 PDFs with selectable text from opendataloader-bench. Vision has the highest score in all four accuracy measures shown here. Standard processes the same set at 0.004 seconds per page. The full methodology is in the benchmark documentation.

PRIVACY Your files stay on your computer

The CLI doesn’t upload your PDF or converted output to Nutrient. Usage reports include the conversion mode and input page count, but not file names, paths, document contents, or output. Text you send to an agent is covered by that agent provider’s policies.

Document privacy illustration with a shield and eye icon

STANDARD AND VISION

Standard is free. Vision is priced by page.

The Free plan includes 1,000 Vision pages each month. Choose a larger plan when you need more.

STANDARD

Free without an account

Convert PDFs that contain selectable text to Markdown or text. Standard doesn’t use Vision pages and can continue if Nutrient is unavailable.

VISION

For scans and complex pages

Improve scans, handwriting, formulas, and complex layouts. Vision runs on your computer.


Frequently asked questions

What is the PDF-to-Markdown skill for Claude and Codex?

It’s an agent skill that converts PDFs to structured Markdown on your computer. It preserves headings, lists, tables, and reading order so an agent can work with the result directly.

How do I install the skill?

Run npx skills add pspdfkit-labs/nutrient-skills --skill pdf-to-markdown. The Skills CLI supports Claude Code, Codex, Cursor, Gemini CLI, and other compatible agents.

Does it work with Claude Code and Codex?

Yes. Install the skill with the Skills CLI, or install the PDF-to-Markdown CLI and call it from any agent or script.

How is this different from passing a PDF directly to Claude?

The skill converts the PDF before the model reads it. That gives the model explicit headings, lists, tables, and reading order instead of asking it to infer those relationships from raw PDF data.

What does it cost, and do my PDFs go to Nutrient?

Standard conversion is free. Each input page uses one Vision page from your monthly allowance, and the Free plan includes 1,000 Vision pages per month. Your PDFs and converted files stay on your computer. Usage reports include the conversion mode and input page count, but not file names, paths, document contents, or output.

Can I use it with other agents?

Yes. The skill runs in Claude Code, Codex, or as a standalone CLI. The model provider’s policies apply to text you send to it, while the PDF itself stays on your computer.

What’s the difference between pdf-to-markdown, pdf-to-text, and query?

pdf-to-markdown creates structured Markdown. pdf-to-text creates plain text while preserving columns and spacing. query searches an extracted file and returns the passages that best match a query.

What about scanned or handwritten PDFs?

Use Vision for scans, photographs, handwriting, formulas, and complex layouts. Vision runs on your computer.

Can I work offline?

Standard conversion can continue when Nutrient is unavailable. After Vision connects, it can work offline for one hour and process up to 100 pages before it must reconnect. A document that crosses the 100-page threshold can finish.

MORE FROM NUTRIENT AI

Continue with Nutrient document AI

Use PDF to Markdown alongside Nutrient tools for cloud extraction, editing, and document automation.