RepoAI
Browse MCPs

Pdf Reader MCP Server

PDF documents parsed into structured text, tables, and page-level citations for AI agents.

Data last scanned yesterday · Reviewed today

485 tools★ 884TypeScriptMIT

Overview

Citra provides AI agents with a document twin that includes structured text, table data, and OCR output. By maintaining provenance through page-level citations, it allows agents to reference specific table cells or document sections rather than relying on standard text extraction which often loses formatting. Designed as a local-first PDF processing tool, it offers capabilities via SDK, CLI, or MCP. It focuses on delivering accurate visual evidence and structure, aimed at reducing errors and hallucinations in document-based AI workflows.

Our verdict

With a security score of 48/100, this server is classified as high-risk. While it features zero direct dependencies and recent active development, critical security vectors like read-only mode, authentication, and OAuth support have not yet been editorially reviewed. It is suitable for local, non-sensitive document processing by informed developers, but should be avoided in environments requiring verified, enterprise-grade security hardening until a formal review process occurs.

  • Zero direct dependencies minimize supply-chain surface area.
  • Provides structured table data, OCR, and page-level provenance.
  • Maintains an active development cycle with recent commits.
  • Licensed under the permissive MIT license.
  • Security score of 48/100 indicates significant unreviewed risk.
  • Lacks external editorial review of its internal tools.
  • Read-only and authentication modes remain unverified.

Setup

Claude Desktop
{
  "mcpServers": {
    "pdf-reader-mcp": {
      "args": [
        "-y",
        "@sylphx/pdf-reader-mcp"
      ],
      "command": "npx"
    }
  }
}

Paste into ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows), then restart Claude Desktop. Some servers require API keys or tokens as environment variables — check the project's README and add an "env" object if needed.

  • No read-only mode — running this grants full read/write access.
  • No built-in authentication — anyone with access to this config can invoke every tool it exposes.

Tools

ToolDescriptionRisk
read_pdfPrimary structured PDF read (text, tables, map, citations).low
search_pdfFind page + snippet matches before deep reading.low
pdf_evidenceCrops, renders, inspect, focused evidence ops.low
render_pagePage render evidence.low
ocr_pagesOCR evidence for scanned pages.low

Compatibility

ClientLocalDockerRemoteRead-only
Claude Desktop
Cursor
VS Code
Windsurf
ChatGPT

Frequently asked questions

›How do I install the Pdf Reader MCP server?

You can install it globally via npm using the command 'npm install -g @sylphx/pdf-reader-mcp'.

›Does this server run on all platforms?

Yes, it provides platform-specific native packages for macOS (arm64/x64), Linux (x64/arm64), and Windows (x64). The installation process automatically selects the correct native package for your host system.

›What happens if a native package is missing for my platform?

The server will fail closed to prevent silent engine switching or unexpected behavior.

›Can I use this without an MCP client?

Yes, you can use the TypeScript SDK directly in your applications or use the CLI via the 'pdf-reader-mcp' binary.

›What primary tools does the server provide to AI agents?

The server provides three core tools: 'read_pdf' for structured text and tables, 'search_pdf' for finding snippets, and 'pdf_evidence' for focused document operations.

›How should I configure the server for Claude Desktop or other MCP clients?

Add an entry to your MCP configuration file under 'mcpServers' using 'npx' as the command and ['@sylphx/pdf-reader-mcp'] as the arguments.

Changelog

Badge

Pdf Reader MCP Server security score, rated on RepoAI

Maintain this server? Add the live badge to your README.