PDF Tools MCP Server

English | 中文

中文

一个基于 FastMCP 的 PDF 读取和操作工具服务器，支持从 PDF 文件的指定页面范围提取文本内容。

功能特性

• 📄 读取 PDF 文件指定页面范围的内容

• 🔢 支持起始和结束页面参数（包含范围）

• 🛡️ 自动处理无效页码（负数、超出范围等）

• 📊 获取 PDF 文件的基本信息

• 🔗 合并多个 PDF 文件

• ✂️ 提取 PDF 的特定页面

• 🔍 正则表达式搜索功能，支持分页查看结果

• 🌐 URL 支持 - 支持直接从 URL 读取和操作 PDF 文件

• 💾 智能缓存机制，相同 URL 的 PDF 自动复用临时文件

安装

从 PyPI 安装

uv add pdf-tools-mcp

如果 uv add 遇到依赖冲突，建议使用：

uvx tool install pdf-tools-mcp

从源码安装

git clone https://github.com/yourusername/pdf-tools-mcp.git
cd pdf-tools-mcp
uv sync

使用方法

与 Claude Desktop 集成

添加到你的 ~/.config/claude/claude_desktop_config.json (Linux/Windows) 或 ~/Library/Application Support/Claude/claude_desktop_config.json (macOS):

开发/未发布版本配置

{
  "mcpServers": {
    "pdf-tools-mcp": {
      "command": "uv",
      "args": [
        "--directory",
        "<path/to/the/repo>/pdf-tools-mcp",
        "run",
        "pdf-tools-mcp",
        "--workspace_path",
        "</your/workspace/directory>",
        "--tempfile_dir",
        "</your/temp/directory>"
      ]
    }
  }
}

已发布版本配置

{
  "mcpServers": {
    "pdf-tools-mcp": {
      "command": "uvx",
      "args": [
        "pdf-tools-mcp",
        "--workspace_path",
        "</your/workspace/directory>",
        "--tempfile_dir",
        "</your/temp/directory>"
      ]
    }
  }
}

注意: 出于安全考虑，此工具只能访问指定工作目录(--workspace_path)内的文件，无法访问工作目录之外的文件。

如果配置后无法正常工作或在UI中无法显示，请通过 uv cache clean 清除缓存。

作为命令行工具

# 基本使用
pdf-tools-mcp

# 指定工作目录和临时文件目录
pdf-tools-mcp --workspace_path /path/to/workspace --tempfile_dir /path/to/temp

作为 Python 包

from pdf_tools_mcp import read_pdf_pages, get_pdf_info, merge_pdfs, extract_pdf_pages

# 读取 PDF 页面（支持 URL）
result = await read_pdf_pages("https://example.com/document.pdf", 1, 5)

# 获取 PDF 信息（支持 URL）
info = await get_pdf_info("document.pdf")

# 合并 PDF 文件（支持 URL 和本地文件混合）
result = await merge_pdfs(["file1.pdf", "https://example.com/file2.pdf"], "merged.pdf")

# 提取特定页面
result = await extract_pdf_pages("source.pdf", [1, 3, 5], "extracted.pdf")

主要工具函数

1. read_pdf_pages

读取 PDF 文件指定页面范围的内容

参数:

• pdf_file_path (str): PDF 文件路径或 URL

• start_page (int, 默认 1): 起始页码

• end_page (int, 默认 1): 结束页码

URL 支持:

• 支持 http:// 和 https:// 协议的 URL

• 自动下载 PDF 文件到临时目录

• 相同 URL 会复用已下载的文件

• 支持 PDF 文件格式验证

示例:

# 读取本地文件第 1-5 页
result = await read_pdf_pages("document.pdf", 1, 5)

# 读取 URL 中的 PDF 第 10 页
result = await read_pdf_pages("https://example.com/document.pdf", 10, 10)

2. get_pdf_info

获取 PDF 文件的基本信息

参数:

• pdf_file_path (str): PDF 文件路径或 URL

返回信息:

• 总页数

• 标题

• 作者

• 创建者

• 创建日期

3. merge_pdfs

合并多个 PDF 文件

参数:

• pdf_paths (List[str]): 要合并的 PDF 文件路径列表（支持 URL 和本地文件混合）

• output_path (str): 合并后的输出文件路径（必须是本地路径）

4. extract_pdf_pages

从 PDF 中提取特定页面

参数:

• source_path (str): 源 PDF 文件路径或 URL

• page_numbers (List[int]): 要提取的页码列表（从 1 开始）

• output_path (str): 输出文件路径（必须是本地路径）

错误处理

工具自动处理以下情况：

• 负数页码：自动调整为第 1 页

• 超出 PDF 总页数的页码：自动调整为最后一页

• 起始页大于结束页：自动交换

• 文件未找到：返回相应错误信息

• 权限不足：返回相应错误信息

使用示例

# 获取 PDF 信息
info = await get_pdf_info("sample.pdf")
print(info)

# 读取前 3 页
content = await read_pdf_pages("sample.pdf", 1, 3)
print(content)

# 读取最后一页（假设 PDF 有 10 页）
content = await read_pdf_pages("sample.pdf", 10, 10)
print(content)

# 使用 URL 读取 PDF
content = await read_pdf_pages("https://example.com/sample.pdf", 1, 3)
print(content)

# 合并多个 PDF（混合本地文件和 URL）
result = await merge_pdfs([
    "part1.pdf", 
    "https://example.com/part2.pdf", 
    "part3.pdf"
], "complete.pdf")
print(result)

# 从 URL 的 PDF 提取特定页面
result = await extract_pdf_pages("https://example.com/source.pdf", [1, 3, 5, 7], "selected.pdf")
print(result)

注意事项

• 页面范围使用包含区间，即起始页和结束页都包含在内

• 如果指定页面没有文本内容，将被跳过

• 返回结果会显示 PDF 总页数和实际提取的页面范围

• 支持各种语言的 PDF 文档

• 建议一次读取的页面数不超过 50 页，以避免性能问题

• URL 支持说明:

  • 支持 HTTP 和 HTTPS 协议的 URL

  • URL 中的 PDF 会被下载到临时目录（默认：~/.pdf_tools_temp）

  • 相同的 URL 会复用已下载的文件，避免重复下载

  • 下载的文件会进行 PDF 格式验证

  • 输出文件路径（如合并、提取功能）必须是本地路径，不能是 URL

开发

构建

uv build

发布到 PyPI

uv publish

本地开发

# 安装开发依赖
uv sync

# 运行测试
uv run python -m pytest

# 运行服务器
uv run python -m pdf_tools_mcp.server

English

A FastMCP-based PDF reading and manipulation tool server that supports extracting text content from specified page ranges of PDF files.

Features

• 📄 Read content from specified page ranges of PDF files

• 🔢 Support for start and end page parameters (inclusive range)

• 🛡️ Automatic handling of invalid page numbers (negative numbers, out of range, etc.)

• 📊 Get basic information about PDF files

• 🔗 Merge multiple PDF files

• ✂️ Extract specific pages from PDFs

• 🔍 Regular expression search functionality with paginated results

• 🌐 URL Support - Direct support for reading and manipulating PDF files from URLs

• 💾 Smart caching mechanism to automatically reuse temporary files for the same URLs

Installation

Install from PyPI

uv add pdf-tools-mcp

If uv add encounters dependency conflicts, use:

uvx tool install pdf-tools-mcp

Install from source

git clone https://github.com/yourusername/pdf-tools-mcp.git
cd pdf-tools-mcp
uv sync

Usage

Usage with Claude Desktop

Add to your ~/.config/claude/claude_desktop_config.json (Linux/Windows) or ~/Library/Application Support/Claude/claude_desktop_config.json (macOS):

Development/Unpublished Servers Configuration

{
  "mcpServers": {
    "pdf-tools-mcp": {
      "command": "uv",
      "args": [
        "--directory",
        "<path/to/the/repo>/pdf-tools-mcp",
        "run",
        "pdf-tools-mcp",
        "--workspace_path",
        "</your/workspace/directory>",
        "--tempfile_dir",
        "</your/temp/directory>"
      ]
    }
  }
}

Published Servers Configuration

{
  "mcpServers": {
    "pdf-tools-mcp": {
      "command": "uvx",
      "args": [
        "pdf-tools-mcp",
        "--workspace_path",
        "</your/workspace/directory>",
        "--tempfile_dir",
        "</your/temp/directory>"
      ]
    }
  }
}

Note: For security reasons, this tool can only access files within the specified workspace directory (--workspace_path) and cannot access files outside the workspace directory.

In case it's not working or showing in the UI, clear your cache via uv cache clean.

As a command line tool

# Basic usage
pdf-tools-mcp

# Specify workspace directory and temporary file directory
pdf-tools-mcp --workspace_path /path/to/workspace --tempfile_dir /path/to/temp

As a Python package

from pdf_tools_mcp import read_pdf_pages, get_pdf_info, merge_pdfs, extract_pdf_pages

# Read PDF pages (URL support)
result = await read_pdf_pages("https://example.com/document.pdf", 1, 5)

# Get PDF info (URL support)
info = await get_pdf_info("document.pdf")

# Merge PDF files (mixed URLs and local files)
result = await merge_pdfs(["file1.pdf", "https://example.com/file2.pdf"], "merged.pdf")

# Extract specific pages
result = await extract_pdf_pages("source.pdf", [1, 3, 5], "extracted.pdf")

Main Tool Functions

1. read_pdf_pages

Read content from specified page ranges of a PDF file

Parameters:

• pdf_file_path (str): PDF file path or URL

• start_page (int, default 1): Starting page number

• end_page (int, default 1): Ending page number

URL Support:

• Supports http:// and https:// protocol URLs

• Automatically downloads PDF files to temporary directory

• Reuses downloaded files for the same URLs

• Includes PDF file format validation

Example:

# Read pages 1-5 from local file
result = await read_pdf_pages("document.pdf", 1, 5)

# Read page 10 from URL
result = await read_pdf_pages("https://example.com/document.pdf", 10, 10)

2. get_pdf_info

Get basic information about a PDF file

Parameters:

• pdf_file_path (str): PDF file path or URL

Returns:

• Total page count

• Title

• Author

• Creator

• Creation date

3. merge_pdfs

Merge multiple PDF files

Parameters:

• pdf_paths (List[str]): List of PDF file paths to merge (supports mixed URLs and local files)

• output_path (str): Output file path for the merged PDF (must be local path)

4. extract_pdf_pages

Extract specific pages from a PDF

Parameters:

• source_path (str): Source PDF file path or URL

• page_numbers (List[int]): List of page numbers to extract (1-based)

• output_path (str): Output file path (must be local path)

Error Handling

The tool automatically handles the following situations:

• Negative page numbers: automatically adjusted to page 1

• Page numbers exceeding total PDF pages: automatically adjusted to the last page

• Start page greater than end page: automatically swapped

• File not found: returns appropriate error message

• Insufficient permissions: returns appropriate error message

Usage Examples

# Get PDF info
info = await get_pdf_info("sample.pdf")
print(info)

# Read first 3 pages
content = await read_pdf_pages("sample.pdf", 1, 3)
print(content)

# Read last page (assuming PDF has 10 pages)
content = await read_pdf_pages("sample.pdf", 10, 10)
print(content)

# Read PDF from URL
content = await read_pdf_pages("https://example.com/sample.pdf", 1, 3)
print(content)

# Merge multiple PDFs (mixed local files and URLs)
result = await merge_pdfs([
    "part1.pdf", 
    "https://example.com/part2.pdf", 
    "part3.pdf"
], "complete.pdf")
print(result)

# Extract specific pages from URL PDF
result = await extract_pdf_pages("https://example.com/source.pdf", [1, 3, 5, 7], "selected.pdf")
print(result)

Notes

• Page ranges use inclusive intervals, meaning both start and end pages are included

• Pages without text content will be skipped

• Results show total PDF page count and actual extracted page range

• Supports PDF documents in various languages

• Recommended to read no more than 50 pages at a time to avoid performance issues

• URL Support Notes:

  • Supports HTTP and HTTPS protocol URLs

  • PDFs from URLs are downloaded to a temporary directory (default: ~/.pdf_tools_temp)

  • Same URLs reuse downloaded files to avoid duplicate downloads

  • Downloaded files undergo PDF format validation

  • Output file paths (for merge, extract functions) must be local paths, not URLs

Development

Build

uv build

Publish to PyPI

uv publish

Local Development

# Install development dependencies
uv sync

# Run tests
uv run python -m pytest

# Run server
uv run python -m pdf_tools_mcp.server

License

MIT License

Contributing

Issues and Pull Requests are welcome!

Changelog

0.1.4

• 🌐 URL Support: Add support for reading PDF files directly from URLs

  • Support for HTTP and HTTPS protocols

  • Automatic PDF download to temporary directory with UUID naming

  • Smart caching mechanism to reuse downloaded files for same URLs

  • PDF format validation (magic bytes, PyPDF2 compatibility check)

  • URL to temporary file mapping management with JSON storage

• ⚙️ Configuration: Add --tempfile_dir parameter for custom temporary directory

• 🔧 Enhanced Functions: All main functions now support URLs:

  • read_pdf_pages: Read from URLs or local files

  • get_pdf_info: Get info from URLs or local files

  • search_pdf_content: Search in URLs or local files

  • merge_pdfs: Merge mixed URLs and local files

  • extract_pdf_pages: Extract from URLs to local files

• 📚 Documentation: Updated README with URL usage examples and configuration

0.1.3

• Add regex search functionality for PDF content

• Add paginated search results with session management

• Add search navigation (next/prev/go to page)

• Add PDF content caching for improved performance

• Add search session cleanup and memory management

0.1.2

• Initial release

• Support for PDF text extraction

• Support for PDF info retrieval

• Support for PDF merging

• Support for page extraction