pdf-it

A Model Context Protocol (MCP) server and Claude Code skill that turns markdown into PDFs that look like they were made on purpose. Cover page, table of contents, code blocks that hold across page breaks, page-numbered footer. One command from your Claude session to a file you can send to a client.

Why this exists

Anthropic ships a PDF skill for reading and filling PDFs. It does not do composition. Every Claude Code research session ends the same way: you have a wall of useful markdown and no clean way to turn it into a PDF a person would want to read.

pdf-it is the other half. Markdown in, designed PDF out, no Chrome printing, no manual HTML conversion.

A 12-page sample is in examples/designing-ai-agent-uiux.pdf.

Works with

pdf-it is a standard Model Context Protocol server. Any client that supports MCP locally can use it.

| Client | Supported | How to add | |---|---|---| | Claude Desktop (Mac, Windows) | yes | Edit claude_desktop_config.json | | Claude Code (CLI) | yes, plus skill triggers like "save this as PDF" | claude mcp add pdf-it -- npx -y pdf-it-mcp | | Cursor | yes | Edit ~/.cursor/mcp.json | | Cline (VS Code extension) | yes | Edit Cline's MCP settings | | Continue.dev | yes | Add via Continue's MCP config | | Zed | yes | Standard MCP config | | Goose (Block's CLI) | yes | Standard MCP config | | Custom agents via the Anthropic SDK | yes | Wire MCP yourself | | claude.ai (browser) | no | Web does not run local MCP servers | | Claude iOS / Android | no | Mobile does not run local MCP servers |

Hard requirements on any client: Node.js 18 or newer, Google Chrome installed, the client must support MCP.

Install

npm install -g pdf-it-mcp

Or run on demand with npx pdf-it-mcp.

Requirements

• Node.js 18 or newer
• Google Chrome installed (used as the renderer, no extra download)

Configure

Claude Desktop

Edit claude_desktop_config.json:

{
  "mcpServers": {
    "pdf-it": {
      "command": "npx",
      "args": ["-y", "pdf-it-mcp"]
    }
  }
}

Claude Code

claude mcp add pdf-it -- npx -y pdf-it-mcp

Cursor

Add to ~/.cursor/mcp.json:

{
  "mcpServers": {
    "pdf-it": {
      "command": "npx",
      "args": ["-y", "pdf-it-mcp"]
    }
  }
}

Custom Chrome path

If Chrome lives somewhere non-standard:

{
  "mcpServers": {
    "pdf-it": {
      "command": "npx",
      "args": ["-y", "pdf-it-mcp"],
      "env": { "CHROME_PATH": "/path/to/chrome" }
    }
  }
}

Use

In any Claude session connected to the server, ask:

Save this as a PDF

Or any of these phrasings: export as PDF, make a PDF report from this, turn this into a PDF, /pdf. The skill picks up the request and routes it through pdf-it. The output lands in ~/Documents/pdf-it/ by default.

Tools

| Tool | Description | |---|---| | generate_pdf | Convert markdown into a PDF. Accepts a template (research-report or plain), optional title and author for the cover, and an optional output path. | | list_templates | Return the list of available templates with descriptions. |

generate_pdf parameters

| Parameter | Required | Description | |---|---|---| | content | yes | Markdown string to convert | | title | no | Shown on the cover and in the page footer | | author | no | Shown on the cover | | output_path | no | Absolute path for the output. Defaults to ~/Documents/pdf-it/{slug}-{timestamp}.pdf | | template | no | research-report (default) or plain |

Templates

| Name | Description | |---|---| | research-report | Cover page with title, author, and date. Auto-generated table of contents from H1 and H2 headings. Body with proper hierarchy. Footer with title and page number. Best for research, summaries, design docs, reports. | | plain | No cover, no TOC. Dense body content only. Best for short notes and quick exports. |

Skill

This package ships with a Claude Code skill at SKILL.md. Trigger phrases the skill responds to:

• save this as PDF
• export as PDF
• make a PDF report from this
• turn this into a PDF
• generate a PDF
• /pdf

See SKILL.md for the full skill spec.

Examples

The examples folder has a sample generated PDF (designing-ai-agent-uiux.pdf, 12 pages) and the cover and body screenshots used in this README.

Output

By default PDFs are written to ~/Documents/pdf-it/{slug}-{timestamp}.pdf. Pass output_path to override.

Design

System fonts where possible. Inter for body and headings, JetBrains Mono for code, page numbers, and metadata. Pure white paper, near-black ink, neutral hairline borders, no accent colors. Code blocks render without syntax highlighting on purpose: color choices in PDFs age badly.

If you want a different design language, fork the templates and adjust. They live in src/templates/ and are plain HTML and CSS rendered through Puppeteer.

Comparison

| | pdf-it | Anthropic PDF skill | |---|---|---| | Read text from a PDF | no | yes | | Extract tables from a PDF | no | yes | | Fill PDF forms | no | yes | | Merge, split, rotate, watermark | no | yes | | OCR scanned PDFs | no | yes | | Compose a designed PDF from markdown | yes | no | | Cover page, TOC, page numbers | yes | no | | Templates | yes | no |

The two are complementary, not competing.

License

MIT. See LICENSE.

Built by Marsel Bait.