# Pdf MCP

> Pdf MCP is an open-source MCP server for files and documents. Local MCP server to search and read one PDF or a folder of them, with page citations. It runs locally on your machine and is started by your AI client.

- **Kind:** MCP server
- **Pricing:** Open source
- **Platforms:** macOS, Windows, Linux, CLI, self-hosted
- **Repository:** https://github.com/jztan/pdf-mcp
- **MCP Registry name:** io.github.jztan/pdf-mcp
- **Categories:** [files and documents](https://ai-powered.com/categories/files-and-documents)
- **Listing updated:** 2026-11-18

## About

pdf-mcp is built for documents too long to paste into a chat. The agent calls pdf_info to see page count and table of contents, pdf_search to locate content, and pdf_read_pages to read only the pages it needs. Search is hybrid, BM25 keyword plus semantic, and hits come with the page they came from.

The README lists 13 tools, including table extraction, rendering pages as PNG for vision models so text can be checked against what is printed, chart data as (x, y) tables, and OCR for scanned pages through Tesseract. For folders, corpus tools warm, triage and search many PDFs at once, backed by an SQLite cache that persists across restarts. Text returned is treated as untrusted, and hidden text, including injected instructions, is flagged rather than silently passed on.

It runs over stdio, installs from PyPI or as a Claude Desktop .mcpb extension, and is MIT licensed. Reading order handles multi-column and Japanese vertical layouts.

## Features

- **Selective page reading:** Read specific pages or ranges instead of loading the whole document.
- **Hybrid search:** Keyword and semantic search with page and section provenance.
- **Folder corpus tools:** Warm, triage and search a whole folder of PDFs.
- **OCR for scans:** Tesseract OCR on scanned pages, run in parallel.
- **Tables and charts:** Structured table rows and chart data as (x, y) values.
- **Hidden-text flagging:** Reports text a reader cannot see so injected instructions are not passed on silently.

## Good for

- Summarise risk factors from a 200-page annual report
- Search a folder of papers for one topic
- Extract a table from a PDF
- OCR a scanned document
- Check extracted text against a rendered page

## Install

### Claude Code (Terminal)

```sh
claude mcp add pdf-mcp -- uvx pdf-mcp
```

### Claude Desktop (claude_desktop_config.json)

```json
{
  "mcpServers": {
    "pdf-mcp": {
      "command": "uvx",
      "args": [
        "pdf-mcp"
      ]
    }
  }
}
```

### Cursor (~/.cursor/mcp.json)

```json
{
  "mcpServers": {
    "pdf-mcp": {
      "command": "uvx",
      "args": [
        "pdf-mcp"
      ]
    }
  }
}
```

### VS Code (.vscode/mcp.json)

```json
{
  "servers": {
    "pdf-mcp": {
      "type": "stdio",
      "command": "uvx",
      "args": [
        "pdf-mcp"
      ]
    }
  }
}
```

### Windsurf (~/.codeium/windsurf/mcp_config.json)

```json
{
  "mcpServers": {
    "pdf-mcp": {
      "command": "uvx",
      "args": [
        "pdf-mcp"
      ]
    }
  }
}
```

### Other clients (your client's MCP settings (most use this shape))

```json
{
  "mcpServers": {
    "pdf-mcp": {
      "command": "uvx",
      "args": [
        "pdf-mcp"
      ]
    }
  }
}
```

## Tools

- `pdf_info`: Page count, metadata, TOC summary, scanned-page detection
- `pdf_search`: Hybrid keyword and semantic search by page or section
- `pdf_read_pages`: Read specific pages or ranges with OCR, tables and images
- `pdf_read_all`: Read a whole document in one call, byte-capped
- `pdf_get_toc`: Full table of contents
- `pdf_render_pages`: Render pages as PNG for vision models
- `pdf_extract_chart`: Chart data as exact (x, y) tables
- `pdf_corpus_warm`: Warm a folder of PDFs into the cache
- `pdf_corpus_overview`: Per-document triage cards for a folder
- `pdf_corpus_search`: Search across a folder
- `pdf_cache_stats`: Per-document cache breakdown and total size
- `pdf_cache_clear`: Clear expired or all cache entries
- `server_info`: Which optional features and config are active

## FAQ

### What does the cache do?

An SQLite cache avoids re-parsing PDFs and survives restarts; its size cap defaults to 2048 MB.

### How big is the first install?

The Claude Desktop extension downloads about 250 MB of components on first start.

### Do scanned PDFs work?

Yes, through Tesseract OCR; the extension downloads a small English-only version on first use.

### Which Python versions?

Python 3.11 to 3.14.

### What licence?

MIT.

### Is Pdf MCP free?

Yes. Pdf MCP is open source, so you can use and self-host it for free.

### How do I add Pdf MCP to Claude Code?

Run this command in your terminal: claude mcp add pdf-mcp -- uvx pdf-mcp

### What are the best alternatives to Pdf MCP?

The closest alternatives to Pdf MCP are SSRS2, Csv, Google_sheets, Onedrive and Zovo Spreadsheet Compare.

## Where this comes from

- Description: written from the project's README.
- Install steps: from the official MCP Registry entry.

## Alternatives

- [SSRS2](https://ai-powered.com/mcp/ssrs2): SSRS2 is an MCP server for files and documents.
- [Csv](https://ai-powered.com/mcp/csv): Csv is a freemium MCP server for files and documents.
- [Google_sheets](https://ai-powered.com/mcp/google-sheets-3): Google_sheets is a freemium MCP server for files and documents.
- [Onedrive](https://ai-powered.com/mcp/onedrive): Onedrive is a freemium MCP server for files and documents.
- [Zovo Spreadsheet Compare](https://ai-powered.com/mcp/zovo-spreadsheet-compare): Zovo Spreadsheet Compare is an MCP server for files and documents.
- [Video to Audio Converter MCP](https://ai-powered.com/mcp/video-to-audio-converter-mcp): Video to Audio Converter MCP is a free MCP server for design and media.
- [Google_drive](https://ai-powered.com/mcp/google-drive-5): Google_drive is a freemium MCP server for files and documents.
- [Markdown](https://ai-powered.com/mcp/markdown-2): Markdown is a freemium MCP server for files and documents.

---
Source: https://ai-powered.com/mcp/pdf-mcp
